CN107545049A - Image processing method and related product - Google Patents
Image processing method and related product Download PDFInfo
- Publication number
- CN107545049A CN107545049A CN201710713380.7A CN201710713380A CN107545049A CN 107545049 A CN107545049 A CN 107545049A CN 201710713380 A CN201710713380 A CN 201710713380A CN 107545049 A CN107545049 A CN 107545049A
- Authority
- CN
- China
- Prior art keywords
- picture
- target photo
- publish
- illegal
- weight graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the present invention discloses a kind of image processing method and Related product, it is main to make use of cognition similarity comparison technology, it can either identify whether picture possesses copyright and may possess which type of copyright, simultaneously, similar illegal-publish weight graph piece can be recommended to user again, it is easy to user to be replaced use using the similar illegal-publish weight graph piece, so as to help user to evade the application risk that may be brought by copyright problem, lifts the quality of copyright related service.
Description
Technical field
The present invention relates to Internet technical field, and in particular at picture Processing Technique field, more particularly to a kind of picture
Reason method, a kind of picture processing device, a kind of computer-readable storage medium and a kind of service equipment.
Background technology
Picture comparison technology refers to the feature (such as color, fingerprint) that picture is extracted by feature extraction algorithm, then counts
Characteristic similarity between nomogram piece is so as to reaching the purpose of comparison.At present, most picture processing scheme is to utilize figure
Piece comparison technology realizes also have some schemes to be realized on the basis of picture comparison technology using Model Matching in addition, such as:
BoW models (Bag-of-words model, bag of words), machine learning training pattern etc..Above-mentioned existing scheme belongs to
In strict similarity comparison, that is to say, that two similar pictures that comparison result obtains are either from the vision of people still from machine
Think that the two is strict similar from the point of view of vision.For a copyrighted picture, found using this strict similarity comparison
Matching picture be probably greatly very much picture in itself, this matching picture equally exists copyright problem, it is seen then that existing strict similar
The picture processing scheme of comparison is not too much applied to the copyright related service scene for picture.
The content of the invention
The embodiment of the present invention provides a kind of image processing method and Related product, can either realize the identification clothes of picture copyright
Business, and can enough recommend to evade because intentionally or accidentally using copyright picture available for the associated illegal-publish weight graph piece replaced, help
Risk caused by possible, lift practicality.
On the one hand, the embodiment of the present invention provides a kind of image processing method, including:
The characteristic information of pending Target Photo is obtained, the characteristic information includes color characteristic, body region feature
And mark feature;
The characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo is called to confirm the Target Photo
Type, the copyright picture library is associated with the illegal-publish weight graph storehouse;
If identification failure, the attribute information of the Target Photo is obtained, the attribute information includes color attribute, emotion category
Property and text attribute;
According to the attribute information of the Target Photo similar is performed in the copyright picture library and the illegal-publish weight graph storehouse
Match somebody with somebody, the content recommendation to be matched;
Export the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and with it is described
The associated illegal-publish weight graph piece of similar copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
In the above-mentioned technical solutions, for pending Target Photo, analyze its characteristic information first, as color characteristic,
Body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, if identification is lost
The attribute information for just analyzing the Target Photo again is lost, such as color attribute, emotion attribute and text attribute, is performed with reference to picture library similar
Matching recognizes similar copyright picture or illegal-publish weight graph piece to the Target Photo each other to search, and finally exports content recommendation;It is whole
Individual process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its or non-copyright
Picture, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally using copyright
The legal risk that picture is brought, practicality is higher, improves the quality of picture copyright related service.
As a kind of possible embodiment, the copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information and the associated illegal-publish weight graph piece of every copyright picture of every copyright picture;
It is non-that the illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every
The attribute information of copyright picture;
Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition.
In the above-described embodiment, by building copyright picture library and illegal-publish weight graph storehouse in advance, and establish copyright picture with
The incidence relation of illegal-publish weight graph piece, be advantageous to using the picture library built in advance be compared when providing picture copyright related service
To, matching and inquiry, lifted picture processing efficiency.
As a kind of possible embodiment, the color characteristic includes global color's tuning amount and dominant hue vector;
The body region feature includes at least one Feature Descriptor;
The mark feature includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
In the above-described embodiment, because the species of the characteristic information of picture is very more, such as color, texture, shape,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color characteristic, the body region of picture have chosen
Feature and mark feature, conveniently realize that feature based information carries out copyright identification to picture, ensure the reliability of qualification result.
As alternatively possible embodiment, the characteristic information for obtaining pending Target Photo, including:
Travel through the color-values of each pixel of the Target Photo;
According to the color-values of each pixel of the Target Photo, the Target Photo is built using color partition method
Color histogram, the color partition method defines multiple color subregions;
The pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination
Global color's tuning amount;
Dominant hue pixel is extracted from the color histogram of the Target Photo;
The dominant hue pixel quantity in each color subregion is counted, and the target is obtained to statistical result sequential combination
The dominant hue vector of picture.
In the above-described embodiment, the structure by color histogram and analysis, global color's tuning of picture can be obtained
Amount and dominant hue vector, the similarity comparison between picture two-by-two is converted into the likelihood ratio between vector compared with improving at picture
The efficiency of reason.
As another possible embodiment, the characteristic information for obtaining pending Target Photo, in addition to:
At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;
The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
In the above-described embodiment, extracted by algorithm and describe the body region feature of picture, that is, extracted and retouch
The major part of picture has been stated, the similarity comparison between picture two-by-two is converted into the similarity comparison between major part, has been lifted
The efficiency of picture processing.
As another possible embodiment, the characteristic information for obtaining pending Target Photo, in addition to:
Effective tag set is created for the Target Photo;
Judge whether to get the label of the implication for describing the Target Photo expression;
If not getting, the value of the effectively tag set is arranged to empty;
If getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm from institute
Screening obtains at least one effective label in the label got, and at least one effectively label is had into criterion added to described
Label set.
In the above-described embodiment, the implication expressed by picture is described by one or more effectively labels, by having
Effect tag set is converted to the similarity comparison between picture two-by-two the comparison between the set two-by-two of description picture implication, lifting
The efficiency of picture processing.
As another possible embodiment, the Target Photo is identified in the calling copyright picture library and illegal-publish weight graph storehouse
Characteristic information to confirm the type of the Target Photo, including:
The characteristic information of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first matching degree;
Whether judged according to first matching degree in the copyright picture library comprising the version to match with the Target Photo
Weight graph piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;
If not including, then the characteristic information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between the characteristic information of picture;
Judged to whether there is what is with the Target Photo matched in the illegal-publish weight graph storehouse according to second matching degree
Illegal-publish weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;
Fail if being identified in the absence of if.
In the above-described embodiment, the matching degree between picture and Target Photo in picture library is called to enter Target Photo
Row identification, matching degree threshold value can be set according to actual conditions, so as to accurately obtain picture mirror according to the actual requirements
Determine result.
As another possible embodiment, the calculating process of first matching degree or the second matching degree includes:
Calculate matching result A1 between the color characteristic between two pictures, the matching result between body region feature
B1 and effective tag set matching result C1;
Processing is weighted to A1, B1 and C1 according to default Weighted Rule;
The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used between two pictures of expression
Matching degree.
In the above-described embodiment, it is contemplated that any feature (color characteristic, body region feature or mark feature) ratio
To can not all handle all picture/mb-types well, therefore, using the mechanism of weighting point counting, to the matching result of each feature
According to being actually needed setting weight, then the total score after its weighting is calculated, the matching degree between picture is expressed by total score,
So that the result of matching is more fully accurate.
As another possible embodiment, the image processing method also includes:
If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, first mirror are exported
Determine result and comprise at least the copyright picture to match with the Target Photo, and it is associated with the copyright picture to match
Illegal-publish weight graph piece;
If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result is exported, described second
Qualification result comprises at least the illegal-publish weight graph piece to match with the Target Photo.
In the above-described embodiment, qualification result can successfully be exported by identifying, this qualification result includes the copyright figure to match
Piece or the illegal-publish weight graph piece to match, and also recommended it is associated with the copyright picture to match, available for what is be replaced
Illegal-publish weight graph piece, to help user to evade Copyright Risk.
As another possible embodiment, the color attribute includes global color's tuning amount;
The emotion attribute includes emotion phrase set, if the emotion phrase set is non-NULL, represents the emotion word
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set;
The text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text mark
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
In the above-described embodiment, because the species of the attribute information of picture is very more, such as color, description content,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color attribute, the emotion attribute of picture have chosen
And text, convenient realize carry out Similarity matching to picture based on attribute information, ensure the reliability of matching result.
As another possible embodiment, the attribute information for obtaining the Target Photo, including:
The set of emotion phrase and text marking set are created for the Target Photo;
Judge whether to get the target article belonging to the Target Photo;
If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;
If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;
Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple being used to describe the first standby of emotion
Select phrase and multiple for describing the second alternative phrase of implication;
Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from
Screening obtains at least one text marking phrase in multiple second alternative phrases;
At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, by described in extremely
A few text marking phrase is added to the text marking set of the Target Photo.
In the above-described embodiment, capable analysis is dropped into by the summary to article where picture and context segment, can obtained
To the set of emotion phrase and text marking set of picture, the similarity comparison between picture two-by-two is converted into the phase between set
Seemingly compare, improve the efficiency of picture processing.
As another possible embodiment, the attribute information according to the Target Photo is in the copyright picture library
Similarity matching is performed with the illegal-publish weight graph storehouse, the content recommendation to be matched, including:
The attribute information of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first similarity;
Judge whether include the copyright similar to the Target Photo in the copyright picture library according to first similarity
Picture, if obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation push away
Recommend content;
If not including, then the attribute information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between the attribute information of picture;
Judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo non-
Copyright picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
In the above-described embodiment, the similarity between picture and Target Photo in picture library is called to enter Target Photo
Row identification, similarity threshold can be set according to actual conditions, so as to accurately obtain picture phase according to the actual requirements
Like matching result and content recommendation.
As another possible embodiment, the calculating process of first similarity or the second similarity includes:
Calculate analog result A2 between the color attribute between two pictures, the analog result B2 between emotion attribute and
Analog result C2 between text attribute;
Processing is weighted to A2, B2 and C2 according to default Weighted Rule;
The total score S2, the total score S2 of A2, B2 and C2 after weighting processing are calculated between two pictures of expression
Similarity.
In the above-described embodiment, it is contemplated that any attribute Similarity matching can not all handle all picture categories well
Type, therefore, using the mechanism of weighting point counting, to the analog result of each attribute according to being actually needed setting weight, then calculate
Its total score after weighting, expresses the similarity between picture so that result is more fully accurate by total score.
On the other hand, the embodiment of the present invention additionally provides a kind of picture processing device, it may include:
Feature acquiring unit, for obtaining the characteristic information of pending Target Photo, the characteristic information includes color
Feature, body region feature and mark feature;
Unit is identified, for calling copyright picture library and illegal-publish weight graph storehouse to identify the characteristic information of the Target Photo to confirm
The type of the Target Photo, the copyright picture library are associated with the illegal-publish weight graph storehouse;
Attribute acquiring unit, if failing for identifying, obtain the attribute information of the Target Photo, the attribute information bag
Include color attribute, emotion attribute and text attribute;
Matching unit, for the attribute information according to the Target Photo in the copyright picture library and the illegal-publish weight graph storehouse
Middle execution Similarity matching, the content recommendation to be matched;
Recommendation unit, for exporting the content recommendation, the content recommendation includes the version similar to the Target Photo
Weight graph piece and the illegal-publish weight graph piece associated with the similar copyright picture, or including the illegal-publish similar to the Target Photo
Weight graph piece.
In the above-mentioned technical solutions, for pending Target Photo, analyze its characteristic information first, as color characteristic,
Body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, if identification is lost
The attribute information for just analyzing the Target Photo again is lost, such as color attribute, emotion attribute and text attribute, is performed with reference to picture library similar
Matching recognizes similar copyright picture or illegal-publish weight graph piece to the Target Photo each other to search, and finally exports content recommendation;It is whole
Individual process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its or non-copyright
Picture, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally using copyright
The legal risk that picture is brought, practicality is higher, improves the quality of picture copyright related service.
As a kind of possible embodiment, the copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information and the associated illegal-publish weight graph piece of every copyright picture of every copyright picture;
It is non-that the illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every
The attribute information of copyright picture;
Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition.
In the above-described embodiment, by building copyright picture library and illegal-publish weight graph storehouse in advance, and establish copyright picture with
The incidence relation of illegal-publish weight graph piece, be advantageous to using the picture library built in advance be compared when providing picture copyright related service
To, matching and inquiry, lifted picture processing efficiency.
As alternatively possible embodiment, the color characteristic includes global color's tuning amount and dominant hue vector;
The body region feature includes at least one Feature Descriptor;
The mark feature includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
In the above-described embodiment, because the species of the characteristic information of picture is very more, such as color, texture, shape,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color characteristic, the body region of picture have chosen
Feature and mark feature, conveniently realize that feature based information carries out copyright identification to picture, ensure the reliability of qualification result.
As another possible embodiment, the feature acquiring unit is specifically used for:
Travel through the color-values of each pixel of the Target Photo;
According to the color-values of each pixel of the Target Photo, the Target Photo is built using color partition method
Color histogram, the color partition method defines multiple color subregions;
The pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination
Global color's tuning amount;
Dominant hue pixel is extracted from the color histogram of the Target Photo;And
The dominant hue pixel quantity in each color subregion is counted, and the target is obtained to statistical result sequential combination
The dominant hue vector of picture.
In the above-described embodiment, the structure by color histogram and analysis, global color's tuning of picture can be obtained
Amount and dominant hue vector, the similarity comparison between picture two-by-two is converted into the likelihood ratio between vector compared with improving at picture
The efficiency of reason.
As another possible embodiment, the feature acquiring unit is additionally operable to:
At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;And
The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
In the above-described embodiment, extracted by algorithm and describe the body region feature of picture, that is, extracted and retouch
The major part of picture has been stated, the similarity comparison between picture two-by-two is converted into the similarity comparison between major part, has been lifted
The efficiency of picture processing.
As another possible embodiment, the feature acquiring unit is additionally operable to:
Effective tag set is created for the Target Photo;
Judge whether to get the label of the implication for describing the Target Photo expression;
If not getting, the value of the effectively tag set is arranged to empty;And
If getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm from institute
Screening obtains at least one effective label in the label got, and at least one effectively label is had into criterion added to described
Label set.
In the above-described embodiment, the implication expressed by picture is described by one or more effectively labels, by having
Effect tag set is converted to the similarity comparison between picture two-by-two the comparison between the set two-by-two of description picture implication, lifting
The efficiency of picture processing.
As another possible embodiment, the identification unit is specifically used for:
The characteristic information of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first matching degree;
Whether judged according to first matching degree in the copyright picture library comprising the version to match with the Target Photo
Weight graph piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;
If not including, then the characteristic information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between the characteristic information of picture;
Judged to whether there is what is with the Target Photo matched in the illegal-publish weight graph storehouse according to second matching degree
Illegal-publish weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;And
Fail if being identified in the absence of if.
In the above-described embodiment, the matching degree between picture and Target Photo in picture library is called to enter Target Photo
Row identification, matching degree threshold value can be set according to actual conditions, so as to accurately obtain picture mirror according to the actual requirements
Determine result.
As another possible embodiment, the calculating process of first matching degree or the second matching degree includes:
Calculate matching result A1 between the color characteristic between two pictures, the matching result between body region feature
B1 and effective tag set matching result C1;
Processing is weighted to A1, B1 and C1 according to default Weighted Rule;
The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used between two pictures of expression
Matching degree.
In the above-described embodiment, it is contemplated that any feature (color characteristic, body region feature or mark feature) ratio
To can not all handle all picture/mb-types well, therefore, using the mechanism of weighting point counting, to the matching result of each feature
According to being actually needed setting weight, then the total score after its weighting is calculated, the matching degree between picture is expressed by total score,
So that the result of matching is more fully accurate.
As another possible embodiment, the identification unit is additionally operable to:
If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, first mirror are exported
Determine result and comprise at least the copyright picture to match with the Target Photo, and it is associated with the copyright picture to match
Illegal-publish weight graph piece;
If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result is exported, described second
Qualification result comprises at least the illegal-publish weight graph piece to match with the Target Photo.
In the above-described embodiment, qualification result can successfully be exported by identifying, this qualification result includes the copyright figure to match
Piece or the illegal-publish weight graph piece to match, and also recommended it is associated with the copyright picture to match, available for what is be replaced
Illegal-publish weight graph piece, to help user to evade Copyright Risk.
As another possible embodiment, the color attribute includes global color's tuning amount;
The emotion attribute includes emotion phrase set, if the emotion phrase set is non-NULL, represents the emotion word
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set;
The text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text mark
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
In the above-described embodiment, because the species of the attribute information of picture is very more, such as color, description content,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color attribute, the emotion attribute of picture have chosen
And text, convenient realize carry out Similarity matching to picture based on attribute information, ensure the reliability of matching result.
As another possible embodiment, the attribute acquiring unit is specifically used for:
The set of emotion phrase and text marking set are created for the Target Photo;
Judge whether to get the target article belonging to the Target Photo;
If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;
If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;
Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple being used to describe the first standby of emotion
Select phrase and multiple for describing the second alternative phrase of implication;
Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from
Screening obtains at least one text marking phrase in multiple second alternative phrases;
At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, by described in extremely
A few text marking phrase is added to the text marking set of the Target Photo.
In the above-described embodiment, capable analysis is dropped into by the summary to article where picture and context segment, can obtained
To the set of emotion phrase and text marking set of picture, the similarity comparison between picture two-by-two is converted into the phase between set
Seemingly compare, improve the efficiency of picture processing.
As another possible embodiment, the matching unit is specifically used for:
The attribute information of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first similarity;
Judge whether include the copyright similar to the Target Photo in the copyright picture library according to first similarity
Picture, if obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation push away
Recommend content;
If not including, then the attribute information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between the attribute information of picture;
Judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo non-
Copyright picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
In the above-described embodiment, the similarity between picture and Target Photo in picture library is called to enter Target Photo
Row identification, similarity threshold can be set according to actual conditions, so as to accurately obtain picture phase according to the actual requirements
Like matching result and content recommendation.
As another possible embodiment, the calculating process of first similarity or the second similarity includes:
Calculate analog result A2 between the color attribute between two pictures, the analog result B2 between emotion attribute and
Analog result C2 between text attribute;
Processing is weighted to A2, B2 and C2 according to default Weighted Rule;
The total score S2, the total score S2 of A2, B2 and C2 after weighting processing are calculated between two pictures of expression
Similarity.
In the above-described embodiment, it is contemplated that any attribute Similarity matching can not all handle all picture categories well
Type, therefore, using the mechanism of weighting point counting, to the analog result of each attribute according to being actually needed setting weight, then calculate
Its total score after weighting, expresses the similarity between picture so that result is more fully accurate by total score.
Another further aspect, the embodiment of the present invention additionally provide a kind of computer-readable storage medium, and the computer-readable storage medium is deposited
One or one or more instruction are contained, described one or one or more instruction are suitable to be loaded by processor and perform following steps:
The characteristic information of pending Target Photo is obtained, the characteristic information includes color characteristic, body region feature
And mark feature;
The characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo is called to confirm the Target Photo
Type, the copyright picture library is associated with the illegal-publish weight graph storehouse;
If identification failure, the attribute information of the Target Photo is obtained, the attribute information includes color attribute, emotion category
Property and text attribute;
According to the attribute information of the Target Photo similar is performed in the copyright picture library and the illegal-publish weight graph storehouse
Match somebody with somebody, the content recommendation to be matched;
Export the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and with it is described
The associated illegal-publish weight graph piece of similar copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
In the above-mentioned technical solutions, for pending Target Photo, analyze its characteristic information first, as color characteristic,
Body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, if identification is lost
The attribute information for just analyzing the Target Photo again is lost, such as color attribute, emotion attribute and text attribute, is performed with reference to picture library similar
Matching recognizes similar copyright picture or illegal-publish weight graph piece to the Target Photo each other to search, and finally exports content recommendation;It is whole
Individual process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its or non-copyright
Picture, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally using copyright
The legal risk that picture is brought, practicality is higher, improves the quality of picture copyright related service.
As a kind of possible embodiment, the copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information and the associated illegal-publish weight graph piece of every copyright picture of every copyright picture;
It is non-that the illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every
The attribute information of copyright picture;
Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition.
In the above-described embodiment, by building copyright picture library and illegal-publish weight graph storehouse in advance, and establish copyright picture with
The incidence relation of illegal-publish weight graph piece, be advantageous to using the picture library built in advance be compared when providing picture copyright related service
To, matching and inquiry, lifted picture processing efficiency.
As a kind of possible embodiment, the color characteristic includes global color's tuning amount and dominant hue vector;
The body region feature includes at least one Feature Descriptor;
The mark feature includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
In the above-described embodiment, because the species of the characteristic information of picture is very more, such as color, texture, shape,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color characteristic, the body region of picture have chosen
Feature and mark feature, conveniently realize that feature based information carries out copyright identification to picture, ensure the reliability of qualification result.
As alternatively possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, following steps are specifically performed:
Travel through the color-values of each pixel of the Target Photo;
According to the color-values of each pixel of the Target Photo, the Target Photo is built using color partition method
Color histogram, the color partition method defines multiple color subregions;
The pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination
Global color's tuning amount;
Dominant hue pixel is extracted from the color histogram of the Target Photo;
The dominant hue pixel quantity in each color subregion is counted, and the target is obtained to statistical result sequential combination
The dominant hue vector of picture.
In the above-described embodiment, the structure by color histogram and analysis, global color's tuning of picture can be obtained
Amount and dominant hue vector, the similarity comparison between picture two-by-two is converted into the likelihood ratio between vector compared with improving at picture
The efficiency of reason.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, also execute the following steps:
At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;
The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
In the above-described embodiment, extracted by algorithm and describe the body region feature of picture, that is, extracted and retouch
The major part of picture has been stated, the similarity comparison between picture two-by-two is converted into the similarity comparison between major part, has been lifted
The efficiency of picture processing.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, also execute the following steps:
Effective tag set is created for the Target Photo;
Judge whether to get the label of the implication for describing the Target Photo expression;
If not getting, the value of the effectively tag set is arranged to empty;
If getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm from institute
Screening obtains at least one effective label in the label got, and at least one effectively label is had into criterion added to described
Label set.
In the above-described embodiment, the implication expressed by picture is described by one or more effectively labels, by having
Effect tag set is converted to the similarity comparison between picture two-by-two the comparison between the set two-by-two of description picture implication, lifting
The efficiency of picture processing.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
It is described to call the characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo to confirm the class of the Target Photo
During the step of type, following steps are specifically performed:
The characteristic information of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first matching degree;
Whether judged according to first matching degree in the copyright picture library comprising the version to match with the Target Photo
Weight graph piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;
If not including, then the characteristic information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between the characteristic information of picture;
Judged to whether there is what is with the Target Photo matched in the illegal-publish weight graph storehouse according to second matching degree
Illegal-publish weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;
Fail if being identified in the absence of if.
In the above-described embodiment, the matching degree between picture and Target Photo in picture library is called to enter Target Photo
Row identification, matching degree threshold value can be set according to actual conditions, so as to accurately obtain picture mirror according to the actual requirements
Determine result.
As another possible embodiment, the calculating process of first matching degree or the second matching degree includes:
Calculate matching result A1 between the color characteristic between two pictures, the matching result between body region feature
B1 and effective tag set matching result C1;
Processing is weighted to A1, B1 and C1 according to default Weighted Rule;
The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used between two pictures of expression
Matching degree.
In the above-described embodiment, it is contemplated that any feature (color characteristic, body region feature or mark feature) ratio
To can not all handle all picture/mb-types well, therefore, using the mechanism of weighting point counting, to the matching result of each feature
According to being actually needed setting weight, then the total score after its weighting is calculated, the matching degree between picture is expressed by total score,
So that the result of matching is more fully accurate.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
Following steps:
If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, first mirror are exported
Determine result and comprise at least the copyright picture to match with the Target Photo, and it is associated with the copyright picture to match
Illegal-publish weight graph piece;
If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result is exported, described second
Qualification result comprises at least the illegal-publish weight graph piece to match with the Target Photo.
In the above-described embodiment, qualification result can successfully be exported by identifying, this qualification result includes the copyright figure to match
Piece or the illegal-publish weight graph piece to match, and also recommended it is associated with the copyright picture to match, available for what is be replaced
Illegal-publish weight graph piece, to help user to evade Copyright Risk.
As another possible embodiment, the color attribute includes global color's tuning amount;
The emotion attribute includes emotion phrase set, if the emotion phrase set is non-NULL, represents the emotion word
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set;
The text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text mark
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
In the above-described embodiment, because the species of the attribute information of picture is very more, such as color, description content,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color attribute, the emotion attribute of picture have chosen
And text, convenient realize carry out Similarity matching to picture based on attribute information, ensure the reliability of matching result.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of attribute information of the acquisition Target Photo, following steps are specifically performed:
The set of emotion phrase and text marking set are created for the Target Photo;
Judge whether to get the target article belonging to the Target Photo;
If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;
If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;
Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple being used to describe the first standby of emotion
Select phrase and multiple for describing the second alternative phrase of implication;
Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from
Screening obtains at least one text marking phrase in multiple second alternative phrases;
At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, by described in extremely
A few text marking phrase is added to the text marking set of the Target Photo.
In the above-described embodiment, capable analysis is dropped into by the summary to article where picture and context segment, can obtained
To the set of emotion phrase and text marking set of picture, the similarity comparison between picture two-by-two is converted into the phase between set
Seemingly compare, improve the efficiency of picture processing.
As another possible embodiment, the attribute information according to the Target Photo is in the copyright picture library
Similarity matching is performed with the illegal-publish weight graph storehouse, the content recommendation to be matched, including:
The attribute information of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first similarity;
Judge whether include the copyright similar to the Target Photo in the copyright picture library according to first similarity
Picture, if obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation push away
Recommend content;
If not including, then the attribute information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between the attribute information of picture;
Judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo non-
Copyright picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
In the above-described embodiment, the similarity between picture and Target Photo in picture library is called to enter Target Photo
Row identification, similarity threshold can be set according to actual conditions, so as to accurately obtain picture phase according to the actual requirements
Like matching result and content recommendation.
As another possible embodiment, the calculating process of first similarity or the second similarity includes:
Calculate analog result A2 between the color attribute between two pictures, the analog result B2 between emotion attribute and
Analog result C2 between text attribute;
Processing is weighted to A2, B2 and C2 according to default Weighted Rule;
The total score S2, the total score S2 of A2, B2 and C2 after weighting processing are calculated between two pictures of expression
Similarity.
In the above-described embodiment, it is contemplated that any attribute Similarity matching can not all handle all picture categories well
Type, therefore, using the mechanism of weighting point counting, to the analog result of each attribute according to being actually needed setting weight, then calculate
Its total score after weighting, expresses the similarity between picture so that result is more fully accurate by total score.
Another further aspect, the embodiment of the present invention additionally provide a kind of service equipment, including:
Processor, it is adapted for carrying out one or one or more instruction;And
Computer-readable storage medium, the computer-readable storage medium be stored with one or one or more instruction, described one or
One or more instruction is suitable to be loaded by the processor and perform following steps:
The characteristic information of pending Target Photo is obtained, the characteristic information includes color characteristic, body region feature
And mark feature;
The characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo is called to confirm the Target Photo
Type, the copyright picture library is associated with the illegal-publish weight graph storehouse;
If identification failure, the attribute information of the Target Photo is obtained, the attribute information includes color attribute, emotion category
Property and text attribute;
According to the attribute information of the Target Photo similar is performed in the copyright picture library and the illegal-publish weight graph storehouse
Match somebody with somebody, the content recommendation to be matched;
Export the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and with it is described
The associated illegal-publish weight graph piece of similar copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
In the above-mentioned technical solutions, for pending Target Photo, analyze its characteristic information first, as color characteristic,
Body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, if identification is lost
The attribute information for just analyzing the Target Photo again is lost, such as color attribute, emotion attribute and text attribute, is performed with reference to picture library similar
Matching recognizes similar copyright picture or illegal-publish weight graph piece to the Target Photo each other to search, and finally exports content recommendation;It is whole
Individual process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its or non-copyright
Picture, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally using copyright
The legal risk that picture is brought, practicality is higher, improves the quality of picture copyright related service.
As a kind of possible embodiment, the copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information and the associated illegal-publish weight graph piece of every copyright picture of every copyright picture;
It is non-that the illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every
The attribute information of copyright picture;
Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition.
In the above-described embodiment, by building copyright picture library and illegal-publish weight graph storehouse in advance, and establish copyright picture with
The incidence relation of illegal-publish weight graph piece, be advantageous to using the picture library built in advance be compared when providing picture copyright related service
To, matching and inquiry, lifted picture processing efficiency.
As a kind of possible embodiment, the color characteristic includes global color's tuning amount and dominant hue vector;
The body region feature includes at least one Feature Descriptor;
The mark feature includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
In the above-described embodiment, because the species of the characteristic information of picture is very more, such as color, texture, shape,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color characteristic, the body region of picture have chosen
Feature and mark feature, conveniently realize that feature based information carries out copyright identification to picture, ensure the reliability of qualification result.
As alternatively possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, following steps are specifically performed:
Travel through the color-values of each pixel of the Target Photo;
According to the color-values of each pixel of the Target Photo, the Target Photo is built using color partition method
Color histogram, the color partition method defines multiple color subregions;
The pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination
Global color's tuning amount;
Dominant hue pixel is extracted from the color histogram of the Target Photo;
The dominant hue pixel quantity in each color subregion is counted, and the target is obtained to statistical result sequential combination
The dominant hue vector of picture.
In the above-described embodiment, the structure by color histogram and analysis, global color's tuning of picture can be obtained
Amount and dominant hue vector, the similarity comparison between picture two-by-two is converted into the likelihood ratio between vector compared with improving at picture
The efficiency of reason.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, also execute the following steps:
At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;
The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
In the above-described embodiment, extracted by algorithm and describe the body region feature of picture, that is, extracted and retouch
The major part of picture has been stated, the similarity comparison between picture two-by-two is converted into the similarity comparison between major part, has been lifted
The efficiency of picture processing.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, also execute the following steps:
Effective tag set is created for the Target Photo;
Judge whether to get the label of the implication for describing the Target Photo expression;
If not getting, the value of the effectively tag set is arranged to empty;
If getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm from institute
Screening obtains at least one effective label in the label got, and at least one effectively label is had into criterion added to described
Label set.
In the above-described embodiment, the implication expressed by picture is described by one or more effectively labels, by having
Effect tag set is converted to the similarity comparison between picture two-by-two the comparison between the set two-by-two of description picture implication, lifting
The efficiency of picture processing.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
It is described to call the characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo to confirm the class of the Target Photo
During the step of type, following steps are specifically performed:
The characteristic information of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first matching degree;
Whether judged according to first matching degree in the copyright picture library comprising the version to match with the Target Photo
Weight graph piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;
If not including, then the characteristic information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between the characteristic information of picture;
Judged to whether there is what is with the Target Photo matched in the illegal-publish weight graph storehouse according to second matching degree
Illegal-publish weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;
Fail if being identified in the absence of if.
In the above-described embodiment, the matching degree between picture and Target Photo in picture library is called to enter Target Photo
Row identification, matching degree threshold value can be set according to actual conditions, so as to accurately obtain picture mirror according to the actual requirements
Determine result.
As another possible embodiment, the calculating process of first matching degree or the second matching degree includes:
Calculate matching result A1 between the color characteristic between two pictures, the matching result between body region feature
B1 and effective tag set matching result C1;
Processing is weighted to A1, B1 and C1 according to default Weighted Rule;
The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used between two pictures of expression
Matching degree.
In the above-described embodiment, it is contemplated that any feature (color characteristic, body region feature or mark feature) ratio
To can not all handle all picture/mb-types well, therefore, using the mechanism of weighting point counting, to the matching result of each feature
According to being actually needed setting weight, then the total score after its weighting is calculated, the matching degree between picture is expressed by total score,
So that the result of matching is more fully accurate.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
Following steps:
If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, first mirror are exported
Determine result and comprise at least the copyright picture to match with the Target Photo, and it is associated with the copyright picture to match
Illegal-publish weight graph piece;
If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result is exported, described second
Qualification result comprises at least the illegal-publish weight graph piece to match with the Target Photo.
In the above-described embodiment, qualification result can successfully be exported by identifying, this qualification result includes the copyright figure to match
Piece or the illegal-publish weight graph piece to match, and also recommended it is associated with the copyright picture to match, available for what is be replaced
Illegal-publish weight graph piece, to help user to evade Copyright Risk.
As another possible embodiment, the color attribute includes global color's tuning amount;
The emotion attribute includes emotion phrase set, if the emotion phrase set is non-NULL, represents the emotion word
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set;
The text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text mark
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
In the above-described embodiment, because the species of the attribute information of picture is very more, such as color, description content,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color attribute, the emotion attribute of picture have chosen
And text, convenient realize carry out Similarity matching to picture based on attribute information, ensure the reliability of matching result.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of attribute information of the acquisition Target Photo, following steps are specifically performed:
The set of emotion phrase and text marking set are created for the Target Photo;
Judge whether to get the target article belonging to the Target Photo;
If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;
If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;
Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple being used to describe the first standby of emotion
Select phrase and multiple for describing the second alternative phrase of implication;
Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from
Screening obtains at least one text marking phrase in multiple second alternative phrases;
At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, by described in extremely
A few text marking phrase is added to the text marking set of the Target Photo.
In the above-described embodiment, capable analysis is dropped into by the summary to article where picture and context segment, can obtained
To the set of emotion phrase and text marking set of picture, the similarity comparison between picture two-by-two is converted into the phase between set
Seemingly compare, improve the efficiency of picture processing.
As another possible embodiment, the attribute information according to the Target Photo is in the copyright picture library
Similarity matching is performed with the illegal-publish weight graph storehouse, the content recommendation to be matched, including:
The attribute information of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first similarity;
Judge whether include the copyright similar to the Target Photo in the copyright picture library according to first similarity
Picture, if obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation push away
Recommend content;
If not including, then the attribute information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between the attribute information of picture;
Judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo non-
Copyright picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
In the above-described embodiment, the similarity between picture and Target Photo in picture library is called to enter Target Photo
Row identification, similarity threshold can be set according to actual conditions, so as to accurately obtain picture phase according to the actual requirements
Like matching result and content recommendation.
As another possible embodiment, the calculating process of first similarity or the second similarity includes:
Calculate analog result A2 between the color attribute between two pictures, the analog result B2 between emotion attribute and
Analog result C2 between text attribute;
Processing is weighted to A2, B2 and C2 according to default Weighted Rule;
The total score S2, the total score S2 of A2, B2 and C2 after weighting processing are calculated between two pictures of expression
Similarity.
In the above-described embodiment, it is contemplated that any attribute Similarity matching can not all handle all picture categories well
Type, therefore, using the mechanism of weighting point counting, to the analog result of each attribute according to being actually needed setting weight, then calculate
Its total score after weighting, expresses the similarity between picture so that result is more fully accurate by total score.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is the acquisition flow chart of color characteristic provided in an embodiment of the present invention;
Fig. 2 is the schematic diagram of color histogram provided in an embodiment of the present invention;
Fig. 3 is the acquisition flow chart of body region feature provided in an embodiment of the present invention;
Fig. 4-Fig. 7 is the accompanying drawings of the acquisition process of body region feature provided in an embodiment of the present invention;
Fig. 8 is the schematic diagram of the description label of picture provided in an embodiment of the present invention;
Fig. 9 is the acquisition flow chart of mark feature provided in an embodiment of the present invention;
Figure 10 is the flow chart for the incidence relation that the embodiment of the present invention is established between copyright picture and illegal-publish weight graph piece;
Figure 11 is the acquisition flow chart of emotion attribute provided in an embodiment of the present invention;
Figure 12 is the acquisition flow chart of text attribute provided in an embodiment of the present invention;
Figure 13 is a kind of flow chart of image processing method provided in an embodiment of the present invention;
Figure 14 is the flow chart of another image processing method provided in an embodiment of the present invention;
Figure 15 is a kind of structural representation of picture processing device provided in an embodiment of the present invention;
Figure 16 is a kind of structural representation of service equipment provided in an embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes.
Picture comparison technology refers to, by feature extraction algorithm extraction picture feature, calculate the similarity between picture feature
So as to reach the purpose of comparison;Wherein, common feature extraction algorithm may include but be not limited to:PHash algorithms (Perceptual
Hash algorithm, perceive hash algorithm, abbreviation PHA), SIFT (Scale Invariant Feature Transform,
Scale invariant features transform) algorithm, SURF (Speeded-Up Robust Features, rapid robust feature) algorithm etc..Figure
Piece feature may include but be not limited to:Textural characteristics (such as picture fingerprint), color characteristic, shape facility, spatial relation characteristics.
Picture comparison technology is often applied in the scenes such as picture recognition, picture retrieval, picture identification, specially compares skill using picture
The feature of pending picture is compared art with the feature of known picture, finds similar to pending picture or matches
Picture is known, so as to reach the purpose of identification, retrieval, identification.In the prior art, most picture processing scheme is to utilize figure
Piece comparison technology realizes also have some schemes to be realized on the basis of picture comparison technology using Model Matching in addition, such as BoW
Model, machine learning training pattern etc..Found in practical application, these existing picture processing schemes belong to strict similar
Compare, that is to say, that in the comparison result obtained by existing scheme, two similar pictures either from the vision of people still from
(machine herein can refer to machine:The figure of the picture processing instrument or terminal of APP, plug-in unit etc., server etc.
Piece processing equipment) vision from the point of view of think that the two is strict similar.
Copyright (copyright) is copyright, refers to the power that literature, art, the author of scientific works enjoy to its works
Profit.If the author of certain picture states copyright protection, then the picture is referred to as copyright picture, and the picture needs to obtain awarding for author
Power just can be used, and otherwise invade the copyright of author.On the contrary, the picture for not being declared copyright protection is referred to as illegal-publish weight graph piece.By
It is restricted in the use of copyright picture, therefore is arisen at the historic moment for the copyright related service of picture, as copyright is identified and is recommended
Service.Copyright is identified and recommendation service refers to identify whether pending picture belongs to copyright figure by means such as picture comparison technologies
Piece, and recommend when being identified as copyright picture available for the illegal-publish weight graph piece being replaced, so as to help to a certain extent
Evade the legal risk caused by intentionally or accidentally using copyright picture possible.It has been observed that due to existing picture processing scheme
Strict similarity comparison is belonged to, the picture that is similar or matching found using this strict similarity comparison is probably very much greatly
This pending picture is in itself, then the pending picture and the picture that is similar or matching found have copyright
Problem, this can not just provide the user recommendation service, can do nothing to help the application risk evaded and brought by copyright problem.It is it can be seen that existing
The picture processing scheme for the strict similarity comparison having not too much is applied to the copyright related service scene for picture.When a lot
Wait, user view using the cause of certain photo for no other reason than that interested in the color in picture, or only to certain in picture
Individual part is interested, that is to say, that if user is simply interested in the color of copyright picture, then can search and the copyright
The similar illegal-publish weight graph piece of the color of picture, which is recommended user and is replaced, can evade copyright problem;Equally, if user
It is interested in some part of copyright picture, then searching the illegal-publish weight graph piece similar to the part and recommending user's replacement makes
With can then evade copyright problem.Because strict similarity comparison requires that the characteristic matching degree of the various pieces between two pictures reaches
More than to matching threshold (set according to being actually needed), thus using strict similarity comparison technology can not meet it is this because
The similar and interchangeable demand in part, thus the embodiment of the present invention proposes the definition of cognition similarity comparison.Recognize similarity comparison
It is relative with strict similarity comparison, two similar pictures that it refers to find by picture comparison technology are either from people's
Vision still can not regard as some features (such as color, body region similar, but that rely on picture itself from the vision of machine
Characteristic of field or label etc.) or rely on picture outside some information (the context text message as where picture) it is considered that this two
Opening similar pictures can mutually replace, and when replacing an other similar pictures using a wherein pictures, (be used from people
Family) vision from the point of view of it is this replace be acceptable, then this two pictures belong to cognition it is similar;Such as:Sent out by comparing
Existing two pictures are similar on Color Expression, and the two can be replaced, then it is similar to form cognition for the two;For another example:Pass through comparison
It was found that two pictures are similar on emotional expression, the two can be replaced, then it is similar to form cognition for the two;And for example:Pass through ratio
To finding that the body region part of two pictures is similar, the two can be replaced, then it is similar to form cognition for the two;Etc..
The embodiments of the invention provide a kind of picture processing scheme, can be applied in the copyright related service scene of picture,
Specifically, the embodiment of the present invention at least solves following two problems:One is that user submits Target Photo requirement to obtain copyright phase
Close service, then need to identify whether the Target Photo possesses copyright or search the copyright picture to match with it, it is also necessary to
User recommends to can be used to the similar illegal-publish weight graph piece being replaced, to help to evade the application risk brought by copyright problem.Its
Two are:User submits the URL (Uniform Resource Locator, URL) of article or submits article requirement
Obtain copyright related service, then need to identify whether all pictures possess copyright in article according to the content of article, and be directed to
Possesses the copyright picture that certain picture searching of copyright matches, it is also necessary to recommend to can be used to the similar illegal-publish being replaced to user
Weight graph piece is to help to evade the application risk brought by copyright problem.The picture processing process of the embodiment of the present invention mainly includes:
First, for pending Target Photo, analyze its characteristic information, such as color characteristic, body region feature and mark feature, so
Picture library is called to identify that the Target Photo belongs to copyright picture or illegal-publish weight graph piece afterwards, if identified successfully, if identification Target Photo
Then found available for the similar illegal-publish weight graph piece replaced, the copyright picture that will match to and available for the non-of replacement for copyright picture
Copyright picture exports as qualification result;The illegal-publish weight graph piece conduct that will match to if identifying that Target Photo is illegal-publish weight graph piece
Qualification result exports.Secondly, if identification failure, continues to analyze the attribute information of the Target Photo, such as color attribute, emotion category
Property and text attribute, then call picture library to perform Similarity matching to search the copyright picture similar to the Target Photo or non-copyright
Picture, found if similar copyright picture is found available for the similar illegal-publish weight graph piece replaced, by similar copyright picture and can
Similar illegal-publish weight graph piece for replacement exports as content recommendation;By the similar illegal-publish if similar illegal-publish weight graph piece is found
Weight graph piece exports as content recommendation.Above-mentioned whole picture processing process make use of cognition similarity comparison technology, can either identify
Go out whether Target Photo possesses copyright and may possess which type of copyright, simultaneously as the qualification result of output or recommendation
Content includes recognizes similar illegal-publish weight graph piece to Target Photo each other, then user can with Selection utilization the similar illegal-publish
Weight graph piece is replaced Target Photo and used, so as to help user to evade the application risk that may be brought by copyright problem, lifting
The quality of copyright related service.It is understood that picture processing scheme provided in an embodiment of the present invention can be with compatible existing
Technology, you can, such as can be above-mentioned using cognition phase so that strict similarity comparison technology and cognition similarity comparison technology to be combined
Before being identified like comparison technology, first each picture in pending picture and picture library is carried out using strict similarity comparison technology pre-
Matching, if not matching strict similar copyright picture or illegal-publish weight graph piece, then perform cognition similarity comparison mentioned above
Scheme.
In order to more easily realize picture processing scheme, the embodiment of the present invention carries out the configuration of picture library, figure herein in advance
Storehouse may include copyright picture library and illegal-publish weight graph storehouse.Wherein, copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information of every copyright picture and the associated illegal-publish weight graph piece of every copyright picture.Wrap in illegal-publish weight graph storehouse
Include the attribute information of an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every illegal-publish weight graph piece;Herein
It should be noted that a copyright picture is associated with an illegal-publish weight graph piece to refer to that the two belongs to the similar picture of cognition.Its
In, characteristic information may include:Color characteristic, body region feature and mark feature.Attribute information may include color attribute, emotion
Attribute and text attribute.
It is specific as follows the following detailed description of copyright picture and the configuration process in illegal-publish weight graph storehouse:
First, the acquisition of characteristic information.
1.1) color characteristic, including global color's tuning amount and dominant hue vector.Please also refer to Fig. 1, for any figure
Piece, the acquisition process of its color characteristic may include following steps s11-s15:
S11, travel through the color-values of each pixel of the picture.
The Essential colour that can not be decomposed again in color is referred to as primary colors, and primary colors can synthesize other colors, but other colors are not
The original color of primary colors can be restored;Primary colors herein refers to red (Red), green (Green), blue (Blue) three kinds of colors, three primary colors
All colors can be blended.That is, any number of color is made up of three primary colors, then, traversal picture can obtain
The color-values of each pixel of the picture, the color-values are expressed as (R, G, B), such as:The color-values of pixel one can represent
For (R1, G1, B1), the color-values of pixel two are represented by (R2, G2, B2), by that analogy.
S12, according to the color-values of each pixel of the picture, the color that the picture is built using color partition method is straight
Fang Tu, the color partition method define multiple color subregions.
Color histogram is the color characteristic being widely adopted in picture processing field, and it describes different color and existed
Proportion in whole pictures, and it is not relevant for the locus residing for color.Because arbitrary hue value is made up of three primary colors
, then the single primary colors of each pixel in picture is extracted respectively, can build three kinds of color histograms;As shown in Fig. 2 pin
Color histogram shown in first three figure of right side from top to bottom can be built respectively by extracting single primary colors R, G, B respectively to the picture in left side
Figure.The primary colors of extraction is synthesized again, can obtain a kind of color histogram;Fig. 2 is referred to again, to primary colors R, G, B of extraction
Carry out synthesizing the color histogram that can obtain shown in last figure of right side.For three primary colors, the span of each primary colors is equal
It is worth for totally 256 for [0,255], if each primary colors takes 256 values, then color space can become very large (2563It is individual
Color-values), analyzed and calculated for so big color space to compare and taken time and effort.The embodiment of the present invention uses color
Partition method, color partition method define multiple color subregions, such as:Can be by color span by color partition method
[0,255] is divided into four color subregions or eight color subregions, by taking four color subregions as an example:By span [0,63] point
For the 0th color area, [64,127] are divided into the first color area, and [128,191] are divided into the second color area, and [192,255] are divided into the 3rd color area;
The color space one formed by above-mentioned this color partition method shares 43Individual color-values, be advantageous to lifting to color space
Analysis and the efficiency calculated.After being divided by color partition method to color space, by each pixel in the picture
Each primary colors value belong to the color histogram that the picture to corresponding color area, can be built respectively.
S13, the pixel quantity in each color subregion is counted, and the complete of the picture is obtained to statistical result sequential combination
Office's tone vector.
The color-values of each pixel of picture can fall into a color district's groups and close, such as:Certain pixel i color-values are
(Ri, Gi, Bi), its primary colors RiValue fall into the 0th color area, primary colors GiValue fall into the second color area, primary colors BiValue fall into
3rd color area, then the color-values of the pixel fall into (the 0th color area, the second color area, the 3rd color area) combination in.This step needs
Count assorted district's groups and close included pixel quantity, such as:Color space is divided into four colors according to color zone method
Subregion, then a total of 43=64 kinds of color district's groups are closed, then are counted assorted district's groups conjunction during 64 kinds of color district's groups are closed respectively and included respectively
Pixel quantity, then by count obtained assorted district's groups close in pixel quantity form the vector of one 64 dimension in order,
This vector can be referred to as global color's tuning amount of the picture, available for the integral color for characterizing the picture.
S14, dominant hue pixel is extracted from the color histogram of the picture.
S15, the dominant hue pixel quantity in each color subregion is counted, and the figure is obtained to statistical result sequential combination
The dominant hue vector of piece.
In step s14-s15, the dominant hue pixel of picture can be obtained by analyzing the color histogram of the picture, be specially
Pixel corresponding to the color-values of the overlapping region for the histogram medium-high frequency that gets colors, but count assorted district's groups conjunction again and included
Dominant hue pixel quantity, by count obtained assorted district's groups close in dominant hue pixel quantity form one 64 in order
The vector of dimension, this vector can be referred to as the dominant hue vector of the picture, available for the dominant hue for characterizing the picture.
The embodiment of the present invention can from internet using a large amount of copyright pictures of the acquisitions such as crawler technology and store to
In copyright picture library, copyright picture can be specifically crawled from the website of statement copyright, or copyright picture is specially provided from some
Website in crawl copyright picture, the color characteristic of each copyright picture can be obtained by the s11-s15 that repeats the above steps
And store into copyright picture library.Similarly, substantial amounts of illegal-publish weight graph piece can be crawled from internet and is stored to illegal-publish weight graph storehouse
In, copyright picture can be specifically crawled from the website of without proper notice copyright, or the illegal-publish weight graph increased income specially is provided from some
Illegal-publish weight graph piece is crawled in the website of piece, then the color characteristic of each illegal-publish weight graph piece is obtained by above-mentioned steps s11-s15
And store into illegal-publish weight graph storehouse.
1.2) body region feature, including at least one Feature Descriptor.For any pictures, its body region refers to it
In major part, for example the scenery figure of an Eiffel Tower, main body are exactly Eiffel Tower.In order to extract a figure
The body region feature of piece, some feature extraction algorithms can be used, such as:(Oriented Brief, a kind of feature carry ORB
Take algorithm) algorithm, SIFT algorithms, SURF algorithm etc..The body region of picture by taking ORB algorithms as an example, will be elaborated below
The acquisition flow of feature, please also refer to Fig. 3, the flow comprises the following steps s21-s22:
S21, at least one body region feature pixel is extracted from picture using feature extraction algorithm.
Step s21 is the extraction process that body region feature pixel is carried out for picture.The character pixel of one picture
Put dark in the bright spot that can be understood as in significant pixel in the picture, such as profile point, darker area, brighter areas
Point etc..The purpose of feature extraction algorithm is in the feature pixel in detecting and extracting picture, the spy employed in ORB algorithms
Sign extraction algorithm is FAST algorithms.The core concept of FAST algorithms is:A pixel is chosen with the pixel around it to make
Compare, if selected pixel and most pixel around it are all different, then it is believed that selected by this
Pixel be characterized pixel.Please also refer to Fig. 4, the specific following A-D of extraction process:
A, a pixel P is randomly selected in picture, centered on pixel P, radius is that (R is according to actual warp by R
Test value, it is assumed that R of embodiment of the present invention value is a total of 16 pixels on circle 3);Pixel 1 as shown in Figure 4,
Pixel 2...... and pixel 16.
B, a threshold value t is set, threshold value t can carry out value according to practical experience herein;If the picture of two pixels
The absolute value of the difference of element value is more than t, shows that two pixels differ.It should be noted that the pixel value of pixel refers to
The gray value of pixel.
C, detect whether pixel P is characterized pixel, then need to consider this 16 pixels around pixel P
Point, it is specially:
C1, calculate respectively vertical direction pixel value of the pixel (i.e. pixel 1 and pixel 9) between the P of center it
The absolute value of difference, the i.e. absolute value of the difference of pixel value between calculating pixel 1 and center P, and pixel 9 and center respectively
The absolute value of the difference of pixel value between P;If the absolute value being calculated is all higher than threshold value t, step c2 is carried out;
C2, the pixel (i.e. pixel 1 and pixel 9) for calculating vertical direction and horizontal direction pixel (pixel 5
And pixel 13) pixel value between the P of center difference absolute value, if the absolute value being calculated has at least three to exceed threshold
Value t, then carry out step c3;
The absolute value of the difference of pixel value between c3,16 pixels of calculating and center P, if the absolute value being calculated has
At least nine exceedes threshold value t, it is determined that pixel P is characterized pixel.
By above-mentioned A-C, M feature pixel can be detected in picture, M herein is just whole more than or equal to 1
Number.
D, NMS (Non Maximum Suppression, non-maximum suppress) is carried out to picture, and to choose N, (N is just whole
Number and 1≤N≤M) individual body region feature pixel, be specially:
D1, calculate detected feature pixel FAST score value Score, Score value refer to 16 pixels and
The absolute value summation of the difference of pixel value between center, such as:Feature pixel P FAST score value Scorep=| I1-Ip|+|
I2-Ip|+......+|I16-Ip|, wherein, IpRepresentative feature pixel P pixel value;I1Picture around representative feature pixel P
The pixel value of vegetarian refreshments 1, by that analogy, I16The pixel value of pixel 16 around representative feature pixel P;
The feature pixel quantity that d2, statistics are included in a neighborhood (such as 3x3) centered on feature pixel P, if
Only this feature pixel P in the neighborhood, then selected characteristic pixel P is as body region feature pixel;If the neighbour
Other feature pixels in addition to feature pixel P in domain be present, then choose in the neighborhood FAST in all feature pixels
The maximum feature pixel of value is defined as body region feature pixel.
S22, the Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
Obtain after N number of body region feature pixel, it is necessary to describe these body region feature pixels in some way
Attribute, the description that the output of these attributes is referred to as the body region feature pixel is sub (Feature Descritors).
Description of a body region feature pixel is calculated using BRIEF algorithms in ORB algorithms.The core of BRIEF algorithms
Thought is to choose n (n is positive integer) individual point pair using certain pattern around body region feature pixel, this n point pair
Comparative result in combination as description son.Specifically it is divided into following steps:
E, picture is filtered using Gaussian filter, such as:The Gaussian filter of use may include following parameter:Side
Difference is 2, Gauss window 9*9.Further, noise filtering can also be crossed come further using integral image, the embodiment of the present invention is led to
Noise-sensitive can preferably be solved the problems, such as by crossing Gaussian filter and integral image.
F, choose and take neighborhood window centered on body region feature pixel P, n is randomly selected in the neighborhood window, and (n is
Positive integer) to pixel.For convenience of explanation, n=4, n maximums can get 512 in practical application;It is false please also refer to Fig. 5
If 4 points currently chosen are to being respectively labeled as:P1(X,Y)、P2(X,Y)、P3(X,Y)、P4(X,Y);Define T operations and compare picture
Element value size, T Operation Definitions equation below (1)
In above-mentioned formula (1), IxRepresent pixel X pixel value;IyRepresent pixel Y pixel value.So, it is right respectively
Selected point can obtain the body region feature pixel to carrying out T operations, by obtained result progress coded combination
Feature Descriptor.Assuming that:T (P1 (X, Y))=1, T (P2 (X, Y))=0, T (P3 (X, Y))=0, T (P4 (X, Y))=1, then
Body region feature pixel P Feature Descriptor is 1001.
G, it is using body region feature pixel as original during above-mentioned acquisition Feature Descriptor, during selected point pair
Point, using horizontal direction as transverse axis, coordinate system is established by the longitudinal axis of vertical direction.When picture rotates, coordinate system is constant,
The point that same pattern is taken out is to different, and description being calculated is also different, and this does not meet actual conditions;
Therefore need to re-establish coordinate system, new coordinate system is followed the rotation of picture and is rotated, so taken with model identical
Point out is to uniformity.
By the body region feature pixel that FAST algorithms are extracted does not have direction, in order to realize rotation uniformity,
Also need to calculate the principal direction of body region feature pixel, realized in ORB algorithms using gray scale centroid method, gray scale
Centroid method assumes a skew between the gray scale of feature pixel and barycenter be present, and the vector between gray scale and barycenter can be used for
Represent a direction.Specific calculation formula is as follows:
In above-mentioned formula (2), mpqRepresent with feature pixel P Neighborhood matrix, the barycenter of C representative pictures, (x, y) is adjacent
A pixel in domain, I (x, y) represent the gray value of (x, y);θ represents feature pixel P principal direction.
When calculating Feature Descriptor using BRIEF algorithms, for each body region feature pixel, n is randomly selected
To being rotated after pixel, the anglec of rotation is calculated by above-mentioned formula (2), recycles mode described in F to obtain on this basis
To Feature Descriptor, which solves the problem of rotation uniformity.By above-mentioned E-G, each body region feature can be obtained
The Feature Descriptor of pixel.For example the pictures of ORB algorithm process one are used, result as shown in Figure 6 can be obtained, circle represents
The body region feature pixel identified, the region of dotted line frame sign is the body region of picture.
The body region that the embodiment of the present invention can obtain each copyright picture by the s21-s22 that repeats the above steps is special
Levy and store into copyright picture library.Similarly, the body region that each illegal-publish weight graph piece is obtained by above-mentioned steps s21-s22 is special
Levy and store into illegal-publish weight graph storehouse.
, can be by calculating the most narrow spacing between main body matching characteristic pixel when picture is compared two-by-two
From after filtering out the preferable pixel pair of matching, for example, please also refer to Fig. 7, the Eiffel Tower shot to different angle enters
After row body region feature pixel extracts and compares filtering, obtain preferably matching result, Ai Feier iron can be found referring to Fig. 7
The characteristic point in the region of tower has an accurate matching, and from people visually, the body region of two pictures is exactly angstrom
Fei Er steel towers, then if left side is a copyrighted picture, right side is the picture of a no copyright, and Image to right then may be used
Using one of recommendation results as Image to left.
1.3) feature is marked, includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
When crawling picture, it will usually some extraneous informations of picture are climbed down to come, just include the description of picture among these
Label.Such as Fig. 8 is referred to, multiple description labels when crawling picture shown in Fig. 8 while corresponding to crawling;These labels all may be used
To represent the implication of this pictures to a certain extent, then this pictures can most be expressed by, which being found with regard to needs from these labels, contains
One or more effective labels of justice, these effective labels are included into effective tag set of the picture.Assuming that used in Fig. 8
The label that dotted line frame is marked is effective label, then this pictures effective tag set expression the picture be meant that " my god
Altar-the Hall of Praying for Good Harvest ".
Please also refer to Fig. 9, the embodiment of the present invention is used for obtaining the mark of any pictures using following step s31-s34
Feature (includes effect tag set), specific as follows:
S31, effective tag set is created for picture.
S32, judge whether to get the label of the implication for describing picture expression.
S33, if not getting, the value of the effectively tag set is arranged to empty.
S34, if getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm
Screening obtains at least one effective label from accessed label, and at least one effectively label is had added to described
Imitate tag set.
In step s31-s34, if crawling label, then effective label can be screened by probability statistics algorithm,
If label is not crawled, then effective tag set is just arranged to empty, that is, shows not being indexed to and contains for describing picture
The label of justice.Specifically, the screening process of probability statistics algorithm is as follows:
A) all labels crawled are put into dictionary, does not consider to repeat, be all put into;
B) occurrence number of each label in dictionary is counted;
C) total quantity of label of each label in probability of occurrence=occurrence number/dictionary in dictionary is counted;
D) dictionary is rebuild with the form of (label, probability).All labels crawled for any pictures, to weight
The probability of the label is searched in the dictionary newly built, obtains result such as:Label a, 0.12;Label b, 0.33;Label c,
0.01;...
E) label of picture is ranked up according to the ascending order of probability from low to high, takes preceding K (K is predetermined number) individual label
It is defined as effective label and is added in effective tag set of the picture.The preceding most possible effective expression picture of K label
Implication, i.e. probability are more low more have particularity.
The embodiment of the present invention can obtain the mark feature of each copyright picture simultaneously by the s31-s34 that repeats the above steps
Store into copyright picture library.Similarly, the mark feature of each illegal-publish weight graph piece is obtained by above-mentioned steps s31-s34 and stored
Into copyright picture library.
1.4) incidence relation is established.
From above-mentioned 1.1) -1.3), the characteristic information of each pictures can be shown as (color using a triple table
Feature, body region feature, mark feature), then, please also refer to Figure 10, the embodiment of the present invention establish copyright picture with it is non-
The process of the incidence relation of copyright picture comprises the following steps s41-s44:
S41, any copyright picture is chosen, and obtain the triple of the copyright picture.
S42, choose an illegal-publish weight graph piece successively in illegal-publish weight graph storehouse, obtain the ternary of the illegal-publish weight graph piece of the selection
Group.
S43, the matching result calculated respectively between the color characteristic between two selected pictures are designated as A1, body region
The matching result that matching result between characteristic of field is designated as B1 and marked between feature is designated as C1;According to default Weighted Rule pair
A1, B1 and C1 are weighted processing;The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used to represent
Matching degree between this two pictures.
For the matching of color characteristic:When searching color close picture, picture secondary color tuning amount two-by-two can be calculated
Cosine similarity;When strict similarity comparison, what cosine similarity threshold value can be set is relatively higher, and recognizes the likelihood ratio
To when, what cosine similarity threshold value can be set is relatively lower.It is being version from illegal-publish weight graph storehouse in the embodiment of the present invention
When weighing the illegal-publish weight graph piece that picture searching is associated, first, copyright picture and the global color of illegal-publish weight graph piece between any two are calculated
The cosine similarity of tuning amount, those the illegal-publish weight graph pieces of screening more than threshold value.Similarity threshold can be set to 85%, so may be used
To allow picture similar in more multicolour to be selected in, the picture of wide of the mark largely can be also introduced, but this threshold value pair
There is adaptation effect well in scenery picture;Next, calculates the cosine similarity of dominant hue vector between picture two-by-two, and screening is more than
Those illegal-publish weight graph pieces of threshold value.Threshold value can be now set to 95%, so ensure that the picture of wide of the mark does not appear in knot
In fruit collection.
For the matching of body region feature:Assuming that certain body region feature pixel X of copyright picture description is
X:10101011, certain body region feature pixel Y of illegal-publish weight graph piece description are Y:10101010;Set a threshold
Value, such as 80%.When the similarity of X and Y description is more than 90%, X, Y match, and X, Y only have most in this example
Latter position is different, similarity 87.5%, more than 80%;Then X and Y is matching, it is seen that X and Y is carried out into xor operation can
Easily to calculate X and Y matching degree;Body region feature pixel between picture two-by-two is so judged using such scheme successively
Between matching degree, can obtain the matching result of the body region feature between picture two-by-two.
Matching for marking feature:Judge the matching degree between effective tag set of picture two-by-two, such as:Setting one
Individual threshold value is 60%, it is assumed that effective tag set of certain copyright picture includes tri- effective labels of a, b, c, certain illegal-publish weight graph piece
Effective tag set includes tri- effective labels of a, b, d, and the two only has last effective label difference, matching degree 2/3*
100%=66.7% is more than threshold value, then the two is matched.
The matching degree of the characteristic information of two pictures calculates as follows referring to formula (3):
S1=u*A1+v*B1+w*C1(wherein u, v, w represent weight, u+v+w=1, v >=w>u) (3)
S44, Bit-reversed is carried out to illegal-publish weight graph piece according to the order of matching degree from high to low, taking preceding L, (L is positive integer
And be predetermined number) illegal-publish weight graph piece and the copyright picture establish incidence relation.
2nd, the acquisition of attribute information.
When crawling copyright picture and illegal-publish weight graph piece, it will usually while a large amount of high-quality picture and text mixing articles are crawled, lead to
Chang Di, appear in the picture of article or have relation with article content, or picture expresses author and included in article
Certain emotion.Here with two rules:The article of the similar content of description is easier to use similar picture, such as some
Sight spot travel notes;Article with similar emotional expression, picture style also can be close, such as lyric prose.Attribute letter is described below
The acquisition process of breath is as follows:
2.1) color attribute, including global color's tuning amount.
Color can be very good to express the emotion of pictures expression, such as:The picture of one happy emotion of expression, its face
Color is typically bright;The picture of one sad emotion of expression, its color are typically dull.The embodiment of the present invention passes through repetition
Step s11-s13 shown in Fig. 1 can obtain the color attribute of each copyright picture and store into copyright picture library.Similarly, lead to
The color attribute of each illegal-publish weight graph piece can be obtained and store into illegal-publish weight graph storehouse by crossing step s11-s13 shown in Fig. 1.
2.2) emotion attribute, including emotion phrase set, if the emotion phrase set is non-NULL, the emotion word is represented
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set.
Please also refer to Figure 11, the embodiment of the present invention is used for obtaining the feelings of any pictures using following step s51-s57
Feel attribute, it is specific as follows:
S51, emotion phrase set is created for picture.
S52, judge whether to get the article belonging to picture.
S53, if not getting, the value of the emotion phrase set is arranged to empty.
S54, if getting, extract the paragraph up and down of the correspondence position of the full text summary and picture of article in article
Summary.
S55, word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtain multiple being used to describe the of emotion
One alternative phrase.
S56, screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase.
S57, at least one crucial phrase is added to the emotion phrase set of picture.
In step s51-s57, if crawling the affiliated article of picture, then full text summary, the picture of article can be extracted
The context paragraph summary of correspondence position in article, the keyword of expression picture emotion is screened by probability statistics algorithm
Group.If the affiliated article of picture is not crawled, then emotion phrase set is just arranged to empty, that is, shows not to be indexed to be used for
The crucial phrase of picture emotion is described.Specifically, screening process is as follows:
A) text dictionary for word segmentation is prepared.Meaningless word can be included in text dictionary for word segmentation, using this text dictionary for word segmentation
It can facilitate and word segmentation processing is carried out to full text summary and context paragraph summary, weed out meaningless phrase;
B) text emotion dictionary is prepared.Some phrases to show emotion can be included in text emotion dictionary, it is such as (happy, flat
With, detest etc.), using text sentiment dictionary can from full text summary and context paragraph summary participle after all phrases in sieve
Selection sense phrase.Involved probability statistics algorithm in step s31-s34 shown in reference picture 9, can screen to obtain and multiple be used to retouch
The crucial phrase of emotion is stated, these crucial phrases are added to the emotion phrase set of picture.So far, the emotion attribute of picture obtains
Take complete.
The embodiment of the present invention can obtain the emotion attribute of each copyright picture simultaneously by the s51-s57 that repeats the above steps
Store into copyright picture library.Similarly, the emotion attribute of each illegal-publish weight graph piece can be obtained by above-mentioned steps s51-s57 and deposited
Storage is into illegal-publish weight graph storehouse.
2.3) text attribute, including text marking set, if the text marking collection is combined into non-NULL, the text mark is represented
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
Please also refer to Figure 12, the embodiment of the present invention is used for obtaining the feelings of any pictures using following step s61-s67
Feel attribute, it is specific as follows:
S61, text marking set is created for picture.
S62, judge whether to get the article belonging to picture.
S63, if not getting, the value of the text marking set is arranged to empty.
S64, if getting, extract the correspondence position of the full text summary and picture of article in the target article
Upper and lower paragraph summary.
S65, word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtain multiple being used to describe the of implication
Two alternative phrases.
S66, screened using probability statistics algorithm from multiple second alternative phrases and obtain at least one text marking
Phrase.
S67, at least one text marking phrase is added to the text marking set of picture.
In step s61-s67, if crawling the affiliated article of picture, then full text summary, the picture of article can be extracted
The context paragraph summary of correspondence position in article, the text mark of expression picture emotion is screened by probability statistics algorithm
Note phrase.If the affiliated article of picture is not crawled, then text marking set is just arranged to empty, that is, shows not to be indexed to
For describing the text marking phrase of picture implication.Specifically, screening process is as follows:
A) it can be facilitated using text dictionary for word segmentation and word segmentation processing is carried out to full text summary and context paragraph summary, rejected
Fall meaningless phrase;
B) can be made a summary using the label dictionary being related in step s31-s34 shown in Fig. 9 from full text summary and context paragraph
Screen text marking phrase in all phrases after participle, and screen to obtain by probability statistics algorithm and multiple be used to describe picture
The text marking phrase of implication, these text marking phrases are added to the text marking set of picture.So far, the text of picture
Attribute is obtained and finished.
The embodiment of the present invention can obtain the text attribute of each copyright picture simultaneously by the s61-s67 that repeats the above steps
Store into copyright picture library.Similarly, the text attribute of each illegal-publish weight graph piece can be obtained by above-mentioned steps s61-s67 and deposited
Storage is into illegal-publish weight graph storehouse.
So far, the building process of picture library terminates, and the copyright picture library and illegal-publish weight graph storehouse can realize real-time or timing more
Newly;Renewal herein can include:Crawl to a new picture, stored into corresponding picture library, and press from internet
Its characteristic information, attribute information and incidence relation are stored according to said process, or, the picture in picture library carries out the volume such as deleting
During volume operation, synchronized update its characteristic information, attribute information and incidence relation.
Based on foregoing description, the embodiments of the invention provide a kind of image processing method, refers to Figure 13, and this method can wrap
Include following steps S101- steps S105.
S101, obtains the characteristic information of pending Target Photo, and the characteristic information includes color characteristic, body region
Feature and mark feature.
Wherein, in the characteristic information of Target Photo, color characteristic includes global color's tuning amount and dominant hue vector;Body region
Characteristic of field includes at least one Feature Descriptor;Mark feature includes effect tag set, if the effectively tag set is non-
Sky, represent that effective tag set has included at least one effective label for being used to describe Target Photo implication;If effective tally set
Sky is combined into, represents that effective tag set does not include effective label for describing the Target Photo implication.
In the present embodiment, the acquisition process of the color characteristic of Target Photo may include:Travel through each of the Target Photo
The color-values of pixel;According to the color-values of each pixel of the Target Photo, using described in color partition method structure
The color histogram of Target Photo, the color partition method define multiple color subregions;Count in each color subregion
Pixel quantity, and global color's tuning amount of the Target Photo is obtained to statistical result sequential combination;From the Target Photo
Color histogram in extract dominant hue pixel;The dominant hue pixel quantity in each color subregion is counted, and to statistics
As a result sequential combination obtains the dominant hue vector of the Target Photo.This acquisition process may refer to step s11- shown in Fig. 1
S15, it will not be described here.
The acquisition process of the body region feature of Target Photo may include:Using feature extraction algorithm from the Target Photo
The middle at least one body region feature pixel of extraction;Each described body region feature picture of algorithm generation is described using feature
The Feature Descriptor of vegetarian refreshments.This acquisition process can be found in step s21-s22 shown in Fig. 3, will not be described here.
The acquisition process of the mark feature of Target Photo may include:Effective tag set is created for the Target Photo;Sentence
The disconnected label for whether getting the implication for describing the Target Photo expression;If not getting, by effective label
The value of set is arranged to empty;If getting, the value of the effectively tag set is arranged to non-NULL, and unite using probability
Calculating method is screened from accessed label and obtains at least one effective label, and at least one effectively label is added to
Effective tag set.This acquisition process can be found in step s31-s34 shown in Fig. 9, will not be described here.
S102, the characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo is called to confirm the target
The type of picture, the copyright picture library are associated with the illegal-publish weight graph storehouse.
The purpose of identification is to confirm that Target Photo is copyright picture or illegal-publish weight graph piece.The present embodiment can use figure
Piece comparison technology is identified that picture comparison herein may include strict similarity comparison or cognition similarity comparison, such as:First
Using strict similarity comparison technology, search in picture library with the presence or absence of the copyright picture or illegal-publish weight graph to match with Target Photo
Piece, if there is then identifying successfully, otherwise search and whether included and Target Photo in picture library using cognition similarity comparison technology again
The copyright that matches picture or illegal-publish weight graph piece, if identified comprising if successfully, otherwise identification failure.If identified successfully, show
It is copyright picture or illegal-publish weight graph piece to be able to confirm that Target Photo, then, according to pass if identifying Target Photo for copyright picture
Connection relation is found available for the similar illegal-publish weight graph piece replaced, the copyright picture that will match to and the illegal-publish weight graph available for replacement
Piece exports as qualification result, and can also export and inform that the use of the Target Photo may possess the prompting of Copyright Risk;
The illegal-publish weight graph piece that will match to if identifying that Target Photo is illegal-publish weight graph piece exports as qualification result, and can be with defeated
Go out to inform the prompting of the temporary no copyright risk of use of the Target Photo., whereas if identification failure, shows feature based information
Picture comparison technology can not confirm the type of the Target Photo, it is necessary to start the cognition likelihood ratio convection current subsequently based on attribute information
Journey.
S103, if identification failure, obtain the attribute information of the Target Photo, the attribute information include color attribute,
Emotion attribute and text attribute.
In the attribute information of Target Photo, color attribute includes global color's tuning amount.Emotion attribute includes emotion phrase collection
Close, if the emotion phrase set is non-NULL, represent that at least one description picture emotion has been included in the emotion phrase set
Crucial phrase;If the emotion phrase set is sky, represent that the emotion phrase set is not included for describing picture emotion
Crucial phrase.Text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text marking collection
Close and included at least one text marking phrase for being used to describe picture implication;If the text marking collection is combined into sky, institute is represented
State text marking set and do not include text marking phrase for describing picture implication.
The acquisition process of the color attribute of Target Photo can be found in step s11-s13 shown in Fig. 1, will not be described here.
The acquisition process of the emotion attribute of Target Photo may include:Emotion phrase set is created for the Target Photo;Sentence
The disconnected target article whether got belonging to the Target Photo;If not getting, by the value of the emotion phrase set
It is arranged to empty;If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple use
In the first alternative phrase of description emotion;Screened and obtained at least from multiple first alternative phrases using probability statistics algorithm
One crucial phrase;And at least one crucial phrase is added to the emotion phrase set of the Target Photo.This is obtained
Take process to can be found in step s51-s57 shown in Figure 11, will not be described here.
The acquisition process of the text attribute of Target Photo includes:Text marking set is created for the Target Photo;Judge
Whether the target article Target Photo belonging to is got;If not getting, the value of the text marking set is set
It is set to sky;If getting, extract the target article full text summary and the Target Photo in the target article
The summary of paragraph up and down of correspondence position;Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple be used for
The second alternative phrase of implication is described;Screened using probability statistics algorithm from multiple second alternative phrases obtain it is at least one
Text marking phrase;And at least one text marking phrase is added to the text marking set of the Target Photo.
This process can be found in step s61-s67 shown in Figure 12, will not be described here.
S104, phase is performed in the copyright picture library and the illegal-publish weight graph storehouse according to the attribute information of the Target Photo
Like matching, the content recommendation to be matched.
S105, exports the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and
The illegal-publish weight graph piece associated with the similar copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
In step S104-S105, the purpose of Similarity matching is to search to exist with Target Photo based on attribute information to recognize
Similar copyright picture or illegal-publish weight graph piece.Recognized particular by by the copyright picture in copyright picture library with Target Photo
Similarity comparison, the illegal-publish weight graph piece in illegal-publish weight graph storehouse and Target Photo are subjected to cognition similarity comparison, what is matched pushes away
Recommend content.Found if similar copyright picture is found available for the similar illegal-publish weight graph piece replaced, by similar copyright picture and
Similar illegal-publish weight graph piece available for replacement exports as content recommendation;It is if similar illegal-publish weight graph piece is found that this is similar non-
Copyright picture exports as content recommendation.If it can not also be searched it is understood that being utilized in picture library and recognizing similar technique
It is to similar copyright picture or illegal-publish weight graph piece, then exportable to identify failure prompt message, to remind user can not currently obtain
Copyright qualification result, recommendation service can not be provided.
In above-mentioned image processing method embodiment, for pending Target Photo, its characteristic information is analyzed first, such as
Color characteristic, body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, such as
The attribute information of the Target Photo is just analyzed in fruit identification failure again, such as color attribute, emotion attribute and text attribute, with reference to picture library
Perform Similarity matching and recognize similar copyright picture or illegal-publish weight graph piece each other to the Target Photo to search, finally output is recommended
Content;Whole process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its
Or illegal-publish weight graph piece, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally
The legal risk brought using copyright picture, practicality is higher, improves the quality of picture copyright related service.
The embodiment of the present invention additionally provides another image processing method, refers to Figure 14, and this method may include following step
Rapid S201- steps S216.
S201, obtains the characteristic information of pending Target Photo, and the characteristic information includes color characteristic, body region
Feature and mark feature.
S202, the feature of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
The first matching degree between information.
Whether S203, judge in the copyright picture library to include and match with the Target Photo according to first matching degree
Copyright picture, if the determination result is YES represent include, then be transferred to step S204, identify successfully, confirm that the Target Photo is
Copyright picture.If judged result does not include for no expression, step S205 is transferred to.
S205, the characteristic information of the Target Photo and every illegal-publish weight graph piece in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between characteristic information.
S206, judge to whether there is and the Target Photo phase in the illegal-publish weight graph storehouse according to second matching degree
The illegal-publish weight graph piece matched somebody with somebody, if the determination result is YES represent exist, be then transferred to step S207, identify successfully, confirm the target figure
Piece is illegal-publish weight graph piece.If judged result is not present for no expression, step S208 is transferred to.
S208, fail if being identified in the absence of if.
Step S202-S208 can be the specific refinement step to step S102 shown in Figure 13.It is specifically described qualification process,
The calculating process of first matching degree or the second matching degree therein includes:Calculate between the color characteristic between two pictures
With the matching result B1 and the matching result C1 of effective tag set between result A1, body region feature;According to default weighting
Rule is weighted processing to A1, B1 and C1;The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used
Matching degree between two pictures are represented.This calculating process can be found in the description in step s44 shown in Figure 10, not go to live in the household of one's in-laws on getting married herein
State.
S209, if identification failure, obtain the attribute information of the Target Photo, the attribute information include color attribute,
Emotion attribute and text attribute.
S210, the attribute of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
The first similarity between information.
S211, whether judged according to first similarity in the copyright picture library comprising similar to the Target Photo
Copyright picture, if the determination result is YES represent to include, then be transferred to step S212 obtain the similar copyright picture and with the phase
The illegal-publish weight graph piece being associated like copyright picture generates content recommendation;Step S216 is transferred to afterwards.If judged result is no expression
Do not include, be then transferred to step S213.
S213, the attribute information of the Target Photo and every illegal-publish weight graph piece in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between attribute information.
S214, judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo
Illegal-publish weight graph piece, if the determination result is YES represent exist, be then transferred to step S215, obtain the similar illegal-publish weight graph piece generation
Content recommendation.
Step S210-S214 can be the specific refinement step to step S104 shown in Figure 13, be specifically described Similarity matching
Process, wherein, the calculating process of the first similarity or the second similarity includes:Calculate color attribute between two pictures it
Between analog result A2, the analog result B2 between emotion attribute and the analog result C2 between text attribute;Add according to default
Power rule is weighted processing to A2, B2 and C2;Calculate the total score S2 of A2, B2 and C2 after weighting processing, the total score
S2 is used to represent the similarity between two pictures.
S216, exports the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and
The illegal-publish weight graph piece associated with the similar copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
In the embodiment of above-mentioned image processing method, cognition similarity comparison technology is make use of, mesh can either be identified
Whether piece of marking on a map possesses copyright and may possess which type of copyright, simultaneously as the qualification result or content recommendation of output
Include and recognize similar illegal-publish weight graph piece each other to Target Photo, then user can with Selection utilization the similar illegal-publish weight graph
Piece is replaced Target Photo and used, and so as to help user to evade the application risk that may be brought by copyright problem, lifts copyright
The quality of related service.
The embodiment of the present invention is at least applied to following two scenes, scene one:User submits picture request identification and recommended
Service, Target Photo now are one or more pictures that user is submitted, and pass through above-mentioned Figure 13-embodiment illustrated in fig. 14
Method, can be identified the picture submitted and Similarity matching by picture comparison technology, find the picture submitted with user
Similar copyright picture, informs which kind of copyright the picture that user is submitted has, and where this buys the picture of complete copyright,
And can also recommend user has the no copyright picture of certain closeness relation with the picture submitted.Scene two:User submits text
Chapter URL or article content, Target Photo now is one or more picture in article, by shown in above-mentioned Figure 13-Figure 14
The method of embodiment, all pictures in article can be identified by picture comparison technology, it is indicated which picture is that have
Copyright, and text fragment or entire article content according to where picture, recommend the no copyright that user can be used to replace
Picture.
Based on the description of above-mentioned image processing method embodiment, the embodiment of the invention also discloses a kind of picture processing dress
Put, the picture processing device can be a computer program (including program code), and the computer program can be run on
Individual server formed stand-alone service equipment or by multiple server groups into cluster service equipment in, it can be with fit
Hold to realize the image processing method shown in Figure 13-14 any embodiments;Client herein can be operate in such as PC
The terminal such as (Personal Computer, personal computer), mobile phone, PDA (tablet personal computer), include browser, instant messaging
The application client of application program etc., such as:User can submit picture by client and ask copyright related service,
The picture processing device for then running on service equipment responds and performs image processing method.Figure 15 is referred to, picture processing dress
Put operation such as lower unit:
Feature acquiring unit 101, for obtaining the characteristic information of pending Target Photo, the characteristic information includes face
Color characteristic, body region feature and mark feature;
Identify unit 102, for call copyright picture library and illegal-publish weight graph storehouse identify the characteristic information of the Target Photo with
Confirm the type of the Target Photo, the copyright picture library is associated with the illegal-publish weight graph storehouse;
Attribute acquiring unit 103, if failing for identifying, obtain the attribute information of the Target Photo, the attribute letter
Breath includes color attribute, emotion attribute and text attribute;
Matching unit 104, for the attribute information according to the Target Photo in the copyright picture library and the non-copyright
Similarity matching is performed in picture library, the content recommendation to be matched;
Recommendation unit 105, for exporting the content recommendation, the content recommendation includes similar to the Target Photo
Copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture, or including similar to the Target Photo non-
Copyright picture.
In the above-mentioned technical solutions, for pending Target Photo, analyze its characteristic information first, as color characteristic,
Body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, if identification is lost
The attribute information for just analyzing the Target Photo again is lost, such as color attribute, emotion attribute and text attribute, is performed with reference to picture library similar
Matching recognizes similar copyright picture or illegal-publish weight graph piece to the Target Photo each other to search, and finally exports content recommendation;It is whole
Individual process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its or non-copyright
Picture, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally using copyright
The legal risk that picture is brought, practicality is higher, improves the quality of picture copyright related service.
As a kind of possible embodiment, the copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information and the associated illegal-publish weight graph piece of every copyright picture of every copyright picture;
It is non-that the illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every
The attribute information of copyright picture;
Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition.
In the above-described embodiment, by building copyright picture library and illegal-publish weight graph storehouse in advance, and establish copyright picture with
The incidence relation of illegal-publish weight graph piece, be advantageous to using the picture library built in advance be compared when providing picture copyright related service
To, matching and inquiry, lifted picture processing efficiency.
As alternatively possible embodiment, the color characteristic includes global color's tuning amount and dominant hue vector;
The body region feature includes at least one Feature Descriptor;
The mark feature includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
In the above-described embodiment, because the species of the characteristic information of picture is very more, such as color, texture, shape,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color characteristic, the body region of picture have chosen
Feature and mark feature, conveniently realize that feature based information carries out copyright identification to picture, ensure the reliability of qualification result.
As another possible embodiment, the picture processing device during operation characteristic acquiring unit 101,
Specifically include:
Travel through the color-values of each pixel of the Target Photo;
According to the color-values of each pixel of the Target Photo, the Target Photo is built using color partition method
Color histogram, the color partition method defines multiple color subregions;
The pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination
Global color's tuning amount;
Dominant hue pixel is extracted from the color histogram of the Target Photo;And
The dominant hue pixel quantity in each color subregion is counted, and the target is obtained to statistical result sequential combination
The dominant hue vector of picture.
In the above-described embodiment, the structure by color histogram and analysis, global color's tuning of picture can be obtained
Amount and dominant hue vector, the similarity comparison between picture two-by-two is converted into the likelihood ratio between vector compared with improving at picture
The efficiency of reason.
As another possible embodiment, the picture processing device during operation characteristic acquiring unit 101,
Also include:
At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;And
The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
In the above-described embodiment, extracted by algorithm and describe the body region feature of picture, that is, extracted and retouch
The major part of picture has been stated, the similarity comparison between picture two-by-two is converted into the similarity comparison between major part, has been lifted
The efficiency of picture processing.
As another possible embodiment, the picture processing device during operation characteristic acquiring unit 101,
Also include:
Effective tag set is created for the Target Photo;
Judge whether to get the label of the implication for describing the Target Photo expression;
If not getting, the value of the effectively tag set is arranged to empty;And
If getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm from institute
Screening obtains at least one effective label in the label got, and at least one effectively label is had into criterion added to described
Label set.
In the above-described embodiment, the implication expressed by picture is described by one or more effectively labels, by having
Effect tag set is converted to the similarity comparison between picture two-by-two the comparison between the set two-by-two of description picture implication, lifting
The efficiency of picture processing.
As another possible embodiment, the picture processing device is during unit 102 is identified in operation, specifically
Including:
The characteristic information of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first matching degree;
Whether judged according to first matching degree in the copyright picture library comprising the version to match with the Target Photo
Weight graph piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;
If not including, then the characteristic information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between the characteristic information of picture;
Judged to whether there is what is with the Target Photo matched in the illegal-publish weight graph storehouse according to second matching degree
Illegal-publish weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;And
Fail if being identified in the absence of if.
In the above-described embodiment, the matching degree between picture and Target Photo in picture library is called to enter Target Photo
Row identification, matching degree threshold value can be set according to actual conditions, so as to accurately obtain picture mirror according to the actual requirements
Determine result.
As another possible embodiment, the calculating process of first matching degree or the second matching degree includes:
Calculate matching result A1 between the color characteristic between two pictures, the matching result between body region feature
B1 and effective tag set matching result C1;
Processing is weighted to A1, B1 and C1 according to default Weighted Rule;
The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used between two pictures of expression
Matching degree.
In the above-described embodiment, it is contemplated that any feature (color characteristic, body region feature or mark feature) ratio
To can not all handle all picture/mb-types well, therefore, using the mechanism of weighting point counting, to the matching result of each feature
According to being actually needed setting weight, then the total score after its weighting is calculated, the matching degree between picture is expressed by total score,
So that the result of matching is more fully accurate.
As another possible embodiment, the picture processing device also wraps during unit 102 is identified in operation
Include:
If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, first mirror are exported
Determine result and comprise at least the copyright picture to match with the Target Photo, and it is associated with the copyright picture to match
Illegal-publish weight graph piece;
If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result is exported, described second
Qualification result comprises at least the illegal-publish weight graph piece to match with the Target Photo.
In the above-described embodiment, qualification result can successfully be exported by identifying, this qualification result includes the copyright figure to match
Piece or the illegal-publish weight graph piece to match, and also recommended it is associated with the copyright picture to match, available for what is be replaced
Illegal-publish weight graph piece, to help user to evade Copyright Risk.
As another possible embodiment, the color attribute includes global color's tuning amount;
The emotion attribute includes emotion phrase set, if the emotion phrase set is non-NULL, represents the emotion word
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set;
The text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text mark
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
In the above-described embodiment, because the species of the attribute information of picture is very more, such as color, description content,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color attribute, the emotion attribute of picture have chosen
And text, convenient realize carry out Similarity matching to picture based on attribute information, ensure the reliability of matching result.
As another possible embodiment, the picture processing device during attribute acquiring unit 103 is run,
Specifically include:
The set of emotion phrase and text marking set are created for the Target Photo;
Judge whether to get the target article belonging to the Target Photo;
If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;
If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;
Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple being used to describe the first standby of emotion
Select phrase and multiple for describing the second alternative phrase of implication;
Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from
Screening obtains at least one text marking phrase in multiple second alternative phrases;
At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, by described in extremely
A few text marking phrase is added to the text marking set of the Target Photo.
In the above-described embodiment, capable analysis is dropped into by the summary to article where picture and context segment, can obtained
To the set of emotion phrase and text marking set of picture, the similarity comparison between picture two-by-two is converted into the phase between set
Seemingly compare, improve the efficiency of picture processing.
As another possible embodiment, the picture processing device is during Operating match unit 104, specifically
Including:
The attribute information of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first similarity;
Judge whether include the copyright similar to the Target Photo in the copyright picture library according to first similarity
Picture, if obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation push away
Recommend content;
If not including, then the attribute information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between the attribute information of picture;
Judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo non-
Copyright picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
In the above-described embodiment, the similarity between picture and Target Photo in picture library is called to enter Target Photo
Row identification, similarity threshold can be set according to actual conditions, so as to accurately obtain picture phase according to the actual requirements
Like matching result and content recommendation.
As another possible embodiment, the calculating process of first similarity or the second similarity includes:
Calculate analog result A2 between the color attribute between two pictures, the analog result B2 between emotion attribute and
Analog result C2 between text attribute;
Processing is weighted to A2, B2 and C2 according to default Weighted Rule;
The total score S2, the total score S2 of A2, B2 and C2 after weighting processing are calculated between two pictures of expression
Similarity.
In the above-described embodiment, it is contemplated that any attribute Similarity matching can not all handle all picture categories well
Type, therefore, using the mechanism of weighting point counting, to the analog result of each attribute according to being actually needed setting weight, then calculate
Its total score after weighting, expresses the similarity between picture so that result is more fully accurate by total score.
According to one embodiment of present invention, the step S101-S105 that the image processing method shown in Figure 13 is related to can be with
It is the unit in picture processing device as shown in Figure 15 to perform.For example, the step S101-S105 shown in Figure 13
Can be respectively by the feature acquiring unit 101 shown in Figure 15, identification unit 102, attribute acquiring unit 103, matching unit 104
Performed with output unit 105.
According to another embodiment of the invention, the step S201-S216 that the image processing method shown in Figure 14 is related to can
To be the unit in picture processing device as shown in Figure 15 to perform.For example, step S201 shown in Figure 14,
S202-S208, S209, S210-S215, S216 can be respectively by the feature acquiring unit 101 shown in Figure 15, identification units
102nd, attribute acquiring unit 103, matching unit 104 and output unit 105 perform.
According to another embodiment of the invention, the unit in the picture processing device shown in Figure 15 can respectively or
One or several other units all are merged into form, or some (a little) unit therein can also be split as work(again
Can on smaller multiple units form, this can realize same operation, the technology effect without influenceing embodiments of the invention
The realization of fruit.Said units are the divisions of logic-based function, and in actual applications, the function of a unit can also be by multiple
Unit is realized, or the function of multiple units realized by a unit.In other embodiments of the invention, picture processing fills
Other units can also be included by putting, and in actual applications, these functions can also be assisted to realize by other units, and can be by
Multiple unit cooperations are realized.
According to another embodiment of the invention, can be by including CPU (CPU), random access memory
Transported on the universal computing device of such as computer of the treatment elements such as medium (RAM), read-only storage medium (ROM) and memory element
Computer program (including the program generation for each step that the image processing method that is able to carry out as shown in Figure 13-Figure 14 of row is related to
Code), to construct picture processing appliance arrangement as shown in Figure 15, and to realize the picture processing side of the embodiment of the present invention
Method.The computer program can be recorded in for example on computer readable recording medium storing program for performing, and passes through computer readable recording medium storing program for performing
It is loaded into above-mentioned computing device, and runs wherein.
In the embodiment of above-mentioned picture processing device, cognition similarity comparison technology is make use of, mesh can either be identified
Whether piece of marking on a map possesses copyright and may possess which type of copyright, simultaneously as the qualification result or content recommendation of output
Include and recognize similar illegal-publish weight graph piece each other to Target Photo, then user can with Selection utilization the similar illegal-publish weight graph
Piece is replaced Target Photo and used, and so as to help user to evade the application risk that may be brought by copyright problem, lifts copyright
The quality of related service.
One is additionally provided based on the image processing method shown in above-described embodiment and picture processing device, the embodiment of the present invention
Kind service equipment, the service equipment can be used for the corresponding steps for performing method flow shown in above-mentioned Figure 13-Figure 14.Specific implementation
In, the service equipment described in the embodiment of the present invention can be individual server formed stand-alone service equipment or by multiple services
The cluster service equipment of device composition, in service equipment as shown in figure 16, the quantity of its server can be according to the business need of reality
Increased and decreased;The service equipment of description of the embodiment of the present invention can realize Figure 13-14 any embodiments institute with fit end
The image processing method shown;Client herein can be operate in such as PC (Personal Computer, individual calculus
Machine), mobile phone, the terminal such as PDA (tablet personal computer), include the application client of browser, instant messaging application program etc.,
Such as:User can be submitted picture by client and ask copyright related service, then service equipment is responded and performed at picture
Reason method.Figure 16 is referred to, the internal structure of the service equipment comprises at least processor, user interface and computer storage and is situated between
Matter.Wherein, the processor in service equipment, user interface and computer-readable storage medium can be connected by bus or other modes,
In Figure 16 shown in the embodiment of the present invention exemplified by being connected by bus.
User interface is to realize that user interacts the medium exchanged with information with service equipment, and its concrete embodiment can wrap
The display screen (Display) for output and keyboard (Keyboard) for input etc. are included, it is necessary to illustrate, herein
Keyboard both can be physical keyboard, or touch screen dummy keyboard, can also be the key that is combined with touch screen virtualphase of entity
Disk.It is to be understood, however, that user interface can also include one or more of the other thing of such as mouse and/or control-rod
Manage user interface facilities.Processor (or CPU (Central Processing Unit, central processing unit)) is service equipment
Calculating core and control core, its be adapted for carrying out one or one or more instruction, be particularly adapted to load and perform one or
One or more is instructed so as to realize correlation method flow or corresponding function;Such as:CPU can be used for parsing user to service equipment
Transmitted switching on and shutting down instruction, and control service equipment to carry out switching on and shutting down operation;For another example:CPU can be in service equipment internal junction
All kinds of interaction datas, etc. are transmitted between structure.Computer-readable storage medium (Memory) is the memory device in service equipment, is used for
Deposit program and data.It is understood that computer-readable storage medium herein can both include the built-in storage of service equipment
Medium, naturally it is also possible to the expansion storage medium supported including service equipment.Computer-readable storage medium provides memory space, should
Memory space stores the operating system of service equipment.Also, also housed in the memory space suitable for being loaded by processor
And the one or more than one instructions performed, these instructions can be one or more computer program (including journeys
Sequence code).It should be noted that computer-readable storage medium herein can be high-speed RAM memory or non-unstable
Memory (non-volatile memory), a for example, at least magnetic disk storage;It optionally can also be at least one position
In the computer-readable storage medium away from aforementioned processor.
In embodiments of the present invention, processor loads and performed one or one or more deposited in computer-readable storage medium
Instruction, to realize the corresponding steps of method flow shown in above-mentioned Figure 13-Figure 14;In the specific implementation, in computer-readable storage medium
One or one or more instruction are loaded by processor and perform following steps:
The characteristic information of pending Target Photo is obtained, the characteristic information includes color characteristic, body region feature
And mark feature;
The characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo is called to confirm the Target Photo
Type, the copyright picture library is associated with the illegal-publish weight graph storehouse;
If identification failure, the attribute information of the Target Photo is obtained, the attribute information includes color attribute, emotion category
Property and text attribute;
According to the attribute information of the Target Photo similar is performed in the copyright picture library and the illegal-publish weight graph storehouse
Match somebody with somebody, the content recommendation to be matched;
Export the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and with it is described
The associated illegal-publish weight graph piece of similar copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
In the above-mentioned technical solutions, for pending Target Photo, analyze its characteristic information first, as color characteristic,
Body region feature and mark feature;Recall picture library and identify that it is to belong to copyright picture or illegal-publish weight graph piece, if identification is lost
The attribute information for just analyzing the Target Photo again is lost, such as color attribute, emotion attribute and text attribute, is performed with reference to picture library similar
Matching recognizes similar copyright picture or illegal-publish weight graph piece to the Target Photo each other to search, and finally exports content recommendation;It is whole
Individual process can either identify whether Target Photo possesses copyright, and and can enough finds the copyright picture similar to its or non-copyright
Picture, while can recommend to can be used for the illegal-publish weight graph piece being replaced, so as to help to evade because intentionally or accidentally using copyright
The legal risk that picture is brought, practicality is higher, improves the quality of picture copyright related service.
As a kind of possible embodiment, the copyright picture library includes an at least copyright picture, every copyright picture
Characteristic information, the attribute information and the associated illegal-publish weight graph piece of every copyright picture of every copyright picture;
It is non-that the illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every
The attribute information of copyright picture;
Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition.
In the above-described embodiment, by building copyright picture library and illegal-publish weight graph storehouse in advance, and establish copyright picture with
The incidence relation of illegal-publish weight graph piece, be advantageous to using the picture library built in advance be compared when providing picture copyright related service
To, matching and inquiry, lifted picture processing efficiency.
As a kind of possible embodiment, the color characteristic includes global color's tuning amount and dominant hue vector;
The body region feature includes at least one Feature Descriptor;
The mark feature includes effect tag set, if the effectively tag set is non-NULL, has criterion described in expression
At least one effective label for being used to describe picture implication has been included in label set;If the effectively tag set is sky, institute is represented
State effective tag set and do not include effective label for describing picture implication.
In the above-described embodiment, because the species of the characteristic information of picture is very more, such as color, texture, shape,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color characteristic, the body region of picture have chosen
Feature and mark feature, conveniently realize that feature based information carries out copyright identification to picture, ensure the reliability of qualification result.
As alternatively possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, following steps are specifically performed:
Travel through the color-values of each pixel of the Target Photo;
According to the color-values of each pixel of the Target Photo, the Target Photo is built using color partition method
Color histogram, the color partition method defines multiple color subregions;
The pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination
Global color's tuning amount;
Dominant hue pixel is extracted from the color histogram of the Target Photo;
The dominant hue pixel quantity in each color subregion is counted, and the target is obtained to statistical result sequential combination
The dominant hue vector of picture.
In the above-described embodiment, the structure by color histogram and analysis, global color's tuning of picture can be obtained
Amount and dominant hue vector, the similarity comparison between picture two-by-two is converted into the likelihood ratio between vector compared with improving at picture
The efficiency of reason.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, also execute the following steps:
At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;
The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
In the above-described embodiment, extracted by algorithm and describe the body region feature of picture, that is, extracted and retouch
The major part of picture has been stated, the similarity comparison between picture two-by-two is converted into the similarity comparison between major part, has been lifted
The efficiency of picture processing.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of characteristic information of the pending Target Photo of the acquisition, also execute the following steps:
Effective tag set is created for the Target Photo;
Judge whether to get the label of the implication for describing the Target Photo expression;
If not getting, the value of the effectively tag set is arranged to empty;
If getting, the value of the effectively tag set is arranged to non-NULL, and use probability statistics algorithm from institute
Screening obtains at least one effective label in the label got, and at least one effectively label is had into criterion added to described
Label set.
In the above-described embodiment, the implication expressed by picture is described by one or more effectively labels, by having
Effect tag set is converted to the similarity comparison between picture two-by-two the comparison between the set two-by-two of description picture implication, lifting
The efficiency of picture processing.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
It is described to call the characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo to confirm the class of the Target Photo
During the step of type, following steps are specifically performed:
The characteristic information of the characteristic information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first matching degree;
Whether judged according to first matching degree in the copyright picture library comprising the version to match with the Target Photo
Weight graph piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;
If not including, then the characteristic information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second matching degree between the characteristic information of picture;
Judged to whether there is what is with the Target Photo matched in the illegal-publish weight graph storehouse according to second matching degree
Illegal-publish weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;
Fail if being identified in the absence of if.
In the above-described embodiment, the matching degree between picture and Target Photo in picture library is called to enter Target Photo
Row identification, matching degree threshold value can be set according to actual conditions, so as to accurately obtain picture mirror according to the actual requirements
Determine result.
As another possible embodiment, the calculating process of first matching degree or the second matching degree includes:
Calculate matching result A1 between the color characteristic between two pictures, the matching result between body region feature
B1 and effective tag set matching result C1;
Processing is weighted to A1, B1 and C1 according to default Weighted Rule;
The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used between two pictures of expression
Matching degree.
In the above-described embodiment, it is contemplated that any feature (color characteristic, body region feature or mark feature) ratio
To can not all handle all picture/mb-types well, therefore, using the mechanism of weighting point counting, to the matching result of each feature
According to being actually needed setting weight, then the total score after its weighting is calculated, the matching degree between picture is expressed by total score,
So that the result of matching is more fully accurate.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
Following steps:
If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, first mirror are exported
Determine result and comprise at least the copyright picture to match with the Target Photo, and it is associated with the copyright picture to match
Illegal-publish weight graph piece;
If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result is exported, described second
Qualification result comprises at least the illegal-publish weight graph piece to match with the Target Photo.
In the above-described embodiment, qualification result can successfully be exported by identifying, this qualification result includes the copyright figure to match
Piece or the illegal-publish weight graph piece to match, and also recommended it is associated with the copyright picture to match, available for what is be replaced
Illegal-publish weight graph piece, to help user to evade Copyright Risk.
As another possible embodiment, the color attribute includes global color's tuning amount;
The emotion attribute includes emotion phrase set, if the emotion phrase set is non-NULL, represents the emotion word
The crucial phrase of at least one description picture emotion has been included in group set;If the emotion phrase set is sky, the feelings are represented
The crucial phrase for describing picture emotion is not included in sense phrase set;
The text attribute includes text marking set, if the text marking collection is combined into non-NULL, represents the text mark
At least one text marking phrase for being used to describe picture implication has been included in note set;If the text marking collection is combined into sky, table
Show that the text marking phrase for describing picture implication is not included in the text marking set.
In the above-described embodiment, because the species of the attribute information of picture is very more, such as color, description content,
By testing and analyzing repeatedly, with reference to directly perceived cognition of the vision to picture of people, color attribute, the emotion attribute of picture have chosen
And text, convenient realize carry out Similarity matching to picture based on attribute information, ensure the reliability of matching result.
As another possible embodiment, described one or one or more instruction be suitable to loaded and performed by processor
During the step of attribute information of the acquisition Target Photo, following steps are specifically performed:
The set of emotion phrase and text marking set are created for the Target Photo;
Judge whether to get the target article belonging to the Target Photo;
If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;
If getting, extract the target article full text summary and the Target Photo in the target article
Correspondence position paragraph up and down summary;
Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple being used to describe the first standby of emotion
Select phrase and multiple for describing the second alternative phrase of implication;
Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from
Screening obtains at least one text marking phrase in multiple second alternative phrases;
At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, by described in extremely
A few text marking phrase is added to the text marking set of the Target Photo.
In the above-described embodiment, capable analysis is dropped into by the summary to article where picture and context segment, can obtained
To the set of emotion phrase and text marking set of picture, the similarity comparison between picture two-by-two is converted into the phase between set
Seemingly compare, improve the efficiency of picture processing.
As another possible embodiment, the attribute information according to the Target Photo is in the copyright picture library
Similarity matching is performed with the illegal-publish weight graph storehouse, the content recommendation to be matched, including:
The attribute information of the attribute information and every copyright picture in the copyright picture library of the Target Photo is calculated respectively
Between the first similarity;
Judge whether include the copyright similar to the Target Photo in the copyright picture library according to first similarity
Picture, if obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation push away
Recommend content;
If not including, then the attribute information of the Target Photo and every non-copyright in the illegal-publish weight graph storehouse are calculated respectively
The second similarity between the attribute information of picture;
Judged according to second similarity in the illegal-publish weight graph storehouse with the presence or absence of similar to the Target Photo non-
Copyright picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
In the above-described embodiment, the similarity between picture and Target Photo in picture library is called to enter Target Photo
Row identification, similarity threshold can be set according to actual conditions, so as to accurately obtain picture phase according to the actual requirements
Like matching result and content recommendation.
As another possible embodiment, the calculating process of first similarity or the second similarity includes:
Calculate analog result A2 between the color attribute between two pictures, the analog result B2 between emotion attribute and
Analog result C2 between text attribute;
Processing is weighted to A2, B2 and C2 according to default Weighted Rule;
The total score S2, the total score S2 of A2, B2 and C2 after weighting processing are calculated between two pictures of expression
Similarity.
In the above-described embodiment, it is contemplated that any attribute Similarity matching can not all handle all picture categories well
Type, therefore, using the mechanism of weighting point counting, to the analog result of each attribute according to being actually needed setting weight, then calculate
Its total score after weighting, expresses the similarity between picture so that result is more fully accurate by total score.
In the embodiment of above-mentioned service equipment, cognition similarity comparison technology is make use of, target figure can either be identified
Whether piece possesses copyright and may possess which type of copyright, simultaneously as the qualification result or content recommendation of output wrap
Include and recognized similar illegal-publish weight graph piece each other to Target Photo, then user can be replaced the similar illegal-publish weight graph piece with Selection utilization
Change Target Photo to be used, so as to help user to evade the application risk that may be brought by copyright problem, lifting copyright is related
The quality of service.
It should be appreciated that ought be in this specification and in the appended claims in use, term " comprising " and "comprising" instruction
Described feature, entirety, step, operation, the presence of element and/or component, but it is not precluded from one or more of the other feature, whole
Body, step, operation, element, component and/or its presence or addition for gathering.
It is also understood that the term used in this description of the invention is merely for the sake of the mesh for describing specific embodiment
And be not intended to limit the present invention.As used in description of the invention and appended claims, unless on
Other situations are hereafter clearly indicated, otherwise " one " of singulative, "one" and "the" are intended to include plural form.
It will be further appreciated that the term "and/or" used in description of the invention and appended claims is
Refer to any combinations of one or more of the associated item listed and be possible to combine, and including these combinations.
As used in this specification and in the appended claims, term " if " can be according to context quilt
Be construed to " when ... " or " once " or " in response to determining " or " in response to detecting ".Similarly, phrase " if it is determined that " or
If " detect【Described condition or event】" can be interpreted to mean according to context " once it is determined that " or " in response to true
It is fixed " or " once detect【Described condition or event】" or " in response to detecting【Described condition or event】”.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or the spy for combining the embodiment or example description
Point is contained at least one embodiment or example of the present invention.In this manual, to the schematic representation of above-mentioned term not
Identical embodiment or example must be directed to.Moreover, specific features, structure, material or the feature of description can be with office
Combined in an appropriate manner in one or more embodiments or example.In addition, in the case of not conflicting, the skill of this area
Art personnel can be tied the different embodiments or example and the feature of different embodiments or example described in this specification
Close and combine.
In addition, term " first ", " second " are only used for describing purpose, and it is not intended that instruction or hint relative importance
Or the implicit quantity for indicating indicated technical characteristic.Thus, define " first ", the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the invention, " multiple " are meant that at least two, such as two, three
It is individual etc., unless otherwise specifically defined.
Any process or method described otherwise above description in flow chart or herein is construed as, and represents to include
Module, fragment or the portion of the code of the executable instruction of one or more the step of being used to realize specific logical function or process
Point, and the scope of embodiments of the present invention includes other realization, wherein order that is shown or discussing, bag can not be pressed
Include according to involved function by it is basic simultaneously in the way of or in the opposite order, carry out perform function, this should be by the reality of the present invention
A person of ordinary skill in the field is applied to be understood.
It should be appreciated that each several part of the present invention can be realized with hardware, software, firmware or combinations thereof.Above-mentioned
In embodiment, software that multiple steps or method can be performed in memory and by suitable instruction execution system with storage
Or firmware is realized.If, and in another embodiment, can be with well known in the art for example, realized with hardware
Any one of row technology or their combination are realized:With the logic gates for realizing logic function to data-signal
Discrete logic, have suitable combinational logic gate circuit application specific integrated circuit, programmable gate array (PGA), scene
Programmable gate array (FPGA) etc..In addition, each functional unit in each embodiment of the present invention can be integrated in a processing
In module or unit is individually physically present, can also two or more units be integrated in a module.
Above-mentioned integrated module can both be realized in the form of hardware, can also be realized in the form of software function module.It is described
If integrated module is realized in the form of software function module and as independent production marketing or in use, can also stored
In a computer read/write memory medium.
Above disclosure is only preferred embodiment of present invention, can not limit the right model of the present invention with this certainly
Enclose, therefore the equivalent variations made according to the claims in the present invention, still belong to the scope that the present invention is covered.
Claims (14)
- A kind of 1. image processing method, it is characterised in that including:The characteristic information of pending Target Photo is obtained, the characteristic information includes color characteristic, body region feature and mark Note feature;The characteristic information of copyright picture library and the illegal-publish weight graph storehouse identification Target Photo is called to confirm the class of the Target Photo Type, the copyright picture library are associated with the illegal-publish weight graph storehouse;If identification failure, obtain the attribute information of the Target Photo, the attribute information include color attribute, emotion attribute and Text attribute;Similarity matching is performed in the copyright picture library and the illegal-publish weight graph storehouse according to the attribute information of the Target Photo, obtained To the content recommendation to match;Export the content recommendation, the content recommendation include the copyright picture similar to the Target Photo and to it is described similar The associated illegal-publish weight graph piece of copyright picture, or including the illegal-publish weight graph piece similar to the Target Photo.
- 2. the method as described in claim 1, it is characterised in that the copyright picture library include an at least copyright picture, every The characteristic information of copyright picture, the attribute information of every copyright picture and the associated illegal-publish weight graph piece of every copyright picture;Institute Stating illegal-publish weight graph storehouse includes an at least illegal-publish weight graph piece, the characteristic information of every illegal-publish weight graph piece and every illegal-publish weight graph piece Attribute information;Wherein, a copyright picture is associated with an illegal-publish weight graph piece refers to that the two belongs to the similar picture of cognition;The color characteristic includes global color's tuning amount and dominant hue vector;The body region feature includes at least one feature Description;The mark feature includes effect tag set, if the effectively tag set is non-NULL, represents effective label At least one effective label for being used to describe picture implication has been included in set;If the effectively tag set is sky, described in expression Effective tag set does not include effective label for describing picture implication;The color attribute includes global color's tuning amount;The emotion attribute includes emotion phrase set, if the emotion phrase Collection is combined into non-NULL, represents that the crucial phrase of at least one description picture emotion has been included in the emotion phrase set;If the feelings Sense phrase collection is combined into sky, represents that the crucial phrase for describing picture emotion is not included in the emotion phrase set;The text Attribute includes text marking set, if the text marking collection is combined into non-NULL, represents that the text marking set has been included at least One text marking phrase for being used to describe picture implication;If the text marking collection is combined into sky, the text marking collection is represented Close the text marking phrase do not included for describing picture implication.
- 3. method as claimed in claim 2, it is characterised in that the characteristic information for obtaining pending Target Photo, bag Include:Travel through the color-values of each pixel of the Target Photo;According to the color-values of each pixel of the Target Photo, using the face of the color partition method structure Target Photo Color Histogram, the color partition method define multiple color subregions;The pixel quantity in each color subregion is counted, and the overall situation of the Target Photo is obtained to statistical result sequential combination Tone vector;Dominant hue pixel is extracted from the color histogram of the Target Photo;The dominant hue pixel quantity in each color subregion is counted, and the Target Photo is obtained to statistical result sequential combination Dominant hue vector.
- 4. method as claimed in claim 2, it is characterised in that the characteristic information for obtaining pending Target Photo, bag Include:At least one body region feature pixel is extracted from the Target Photo using feature extraction algorithm;The Feature Descriptor of algorithm generation each body region feature pixel is described using feature.
- 5. method as claimed in claim 2, it is characterised in that the characteristic information for obtaining pending Target Photo, bag Include:Effective tag set is created for the Target Photo;Judge whether to get the label of the implication for describing the Target Photo expression;If not getting, the value of the effectively tag set is arranged to empty;If getting, the value of the effectively tag set is arranged to non-NULL, and using probability statistics algorithm from acquired To label in screening obtain at least one effective label, at least one effective label is added to effective tally set Close.
- 6. the method as described in claim any one of 2-5, it is characterised in that the calling copyright picture library and illegal-publish weight graph storehouse mirror The characteristic information of the fixed Target Photo to confirm the type of the Target Photo, including:Calculate respectively between the characteristic information of the Target Photo and the characteristic information of every copyright picture in the copyright picture library The first matching degree;Whether judged according to first matching degree in the copyright picture library comprising the copyright figure to match with the Target Photo Piece, if being identified comprising if successfully, it is copyright picture to confirm the Target Photo;If not including, then the characteristic information of the Target Photo and every illegal-publish weight graph piece in the illegal-publish weight graph storehouse are calculated respectively Characteristic information between the second matching degree;Judged according to second matching degree in the illegal-publish weight graph storehouse with the presence or absence of the illegal-publish to match with the Target Photo Weight graph piece, if being identified in the presence of if successfully, it is illegal-publish weight graph piece to confirm the Target Photo;Fail if being identified in the absence of if.
- 7. method as claimed in claim 6, it is characterised in that the calculating process bag of first matching degree or the second matching degree Include:Calculate matching result A1 between the color characteristic between two pictures, the matching result B1 between body region feature and The matching result C1 of effective tag set;Processing is weighted to A1, B1 and C1 according to default Weighted Rule;The total score S1 of A1, B1 and C1 after weighting processing are calculated, the total score is used to represent the matching between two pictures Degree.
- 8. method as claimed in claim 6, it is characterised in that also include:If identifying successfully and confirming that the Target Photo is copyright picture, the first qualification result, the first identification knot are exported Fruit comprises at least the copyright picture that matches with the Target Photo, and associated with the copyright picture to match non- Copyright picture;If identifying successfully and confirming that the Target Photo is illegal-publish weight graph piece, the second qualification result, second identification are exported As a result the illegal-publish weight graph piece to match with the Target Photo is comprised at least.
- 9. method as claimed in claim 2, it is characterised in that the attribute information for obtaining the Target Photo, including:The set of emotion phrase and text marking set are created for the Target Photo;Judge whether to get the target article belonging to the Target Photo;If not getting, the value of the emotion phrase set and the text marking set is disposed as sky;If getting, pair of the full text summary and the Target Photo of the target article in the target article is extracted The paragraph up and down of position is answered to make a summary;Word segmentation processing is carried out to full text summary and upper and lower paragraph summary, obtains multiple the first alternative words for being used to describe emotion Group is used to describe the second alternative phrase of implication with multiple;Screened using probability statistics algorithm from multiple first alternative phrases and obtain at least one crucial phrase, and from multiple Screening obtains at least one text marking phrase in the second alternative phrase;At least one crucial phrase is added to the emotion phrase set of the Target Photo, and, at least one by described in Individual text marking phrase is added to the text marking set of the Target Photo.
- 10. method as claimed in claim 9, it is characterised in that the attribute information according to the Target Photo is described Similarity matching is performed in copyright picture library and the illegal-publish weight graph storehouse, the content recommendation to be matched, including:Calculate respectively between the attribute information of the Target Photo and the attribute information of every copyright picture in the copyright picture library The first similarity;Judge whether include the copyright picture similar to the Target Photo in the copyright picture library according to first similarity, If obtained comprising if the similar copyright picture and the illegal-publish weight graph piece associated with the similar copyright picture generation recommend in Hold;If not including, then the attribute information of the Target Photo and every illegal-publish weight graph piece in the illegal-publish weight graph storehouse are calculated respectively Attribute information between the second similarity;Judged to whether there is the non-copyright similar to the Target Photo in the illegal-publish weight graph storehouse according to second similarity Picture, content recommendation is generated if obtaining the similar illegal-publish weight graph piece in the presence of if.
- 11. method as claimed in claim 10, it is characterised in that the calculating process of first similarity or the second similarity Including:Calculate analog result A2 between the color attribute between two pictures, analog result B2 and text between emotion attribute Analog result C2 between attribute;Processing is weighted to A2, B2 and C2 according to default Weighted Rule;Calculate the total score S2 of A2, B2 and C2 after weighting processing, the phase that the total score S2 is used between two pictures of expression Like degree.
- A kind of 12. picture processing device, it is characterised in that including:Feature acquiring unit, for obtaining the characteristic information of pending Target Photo, the characteristic information include color characteristic, Body region feature and mark feature;Unit is identified, for calling copyright picture library and illegal-publish weight graph storehouse to identify that the characteristic information of the Target Photo is described to confirm The type of Target Photo, the copyright picture library are associated with the illegal-publish weight graph storehouse;Attribute acquiring unit, if failing for identifying, the attribute information of the Target Photo is obtained, the attribute information includes face Color attribute, emotion attribute and text attribute;Matching unit, performed for the attribute information according to the Target Photo in the copyright picture library and the illegal-publish weight graph storehouse Similarity matching, the content recommendation to be matched;Recommendation unit, for exporting the content recommendation, the content recommendation includes the copyright figure similar to the Target Photo Piece and the illegal-publish weight graph piece associated with the similar copyright picture, or including the illegal-publish weight graph similar to the Target Photo Piece.
- 13. a kind of computer-readable storage medium, it is characterised in that the computer-readable storage medium is stored with one or one or more refers to Order, described one or one or more instruction be suitable to loaded as processor and perform the picture as described in claim any one of 1-11 Processing method.
- A kind of 14. service equipment, it is characterised in that including:Processor, it is adapted for carrying out one or one or more instruction;AndComputer-readable storage medium, the computer-readable storage medium is stored with one or one or more is instructed, described one or one Instruction is suitable to be loaded as the processor and perform the image processing method as described in claim any one of 1-11 above.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710713380.7A CN107545049B (en) | 2017-08-18 | 2017-08-18 | Picture processing method and related product |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710713380.7A CN107545049B (en) | 2017-08-18 | 2017-08-18 | Picture processing method and related product |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107545049A true CN107545049A (en) | 2018-01-05 |
CN107545049B CN107545049B (en) | 2022-10-14 |
Family
ID=60958361
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710713380.7A Active CN107545049B (en) | 2017-08-18 | 2017-08-18 | Picture processing method and related product |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107545049B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108363487A (en) * | 2018-01-29 | 2018-08-03 | 阿里巴巴集团控股有限公司 | Construction method, dreamland replay method and the device of dreamland playback model |
CN108596867A (en) * | 2018-05-09 | 2018-09-28 | 五邑大学 | A kind of picture bearing calibration and system based on ORB algorithms |
CN109783654A (en) * | 2018-12-11 | 2019-05-21 | 上海紫竹数字创意港有限公司 | A kind of media asset management system and method |
CN109885715A (en) * | 2019-01-25 | 2019-06-14 | 杭州凡闻科技有限公司 | Information processing method and system |
CN110910298A (en) * | 2019-10-23 | 2020-03-24 | 上海突进网络科技有限公司 | Picture processing method and system |
CN111462190A (en) * | 2020-04-20 | 2020-07-28 | 海信集团有限公司 | Intelligent refrigerator and food material input method |
CN113747233A (en) * | 2021-08-24 | 2021-12-03 | 北京达佳互联信息技术有限公司 | Music replacement method and device, electronic equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080008348A1 (en) * | 2006-02-01 | 2008-01-10 | Markmonitor Inc. | Detecting online abuse in images |
US8121989B1 (en) * | 2008-03-07 | 2012-02-21 | Google Inc. | Determining differences between documents |
CN105138867A (en) * | 2014-06-09 | 2015-12-09 | 北大方正集团有限公司 | Method and device for protecting image copyright |
CN106709027A (en) * | 2016-12-28 | 2017-05-24 | 北京小米移动软件有限公司 | Picture recommending method and device |
CN106777177A (en) * | 2016-12-22 | 2017-05-31 | 百度在线网络技术(北京)有限公司 | Search method and device |
-
2017
- 2017-08-18 CN CN201710713380.7A patent/CN107545049B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080008348A1 (en) * | 2006-02-01 | 2008-01-10 | Markmonitor Inc. | Detecting online abuse in images |
US8121989B1 (en) * | 2008-03-07 | 2012-02-21 | Google Inc. | Determining differences between documents |
CN105138867A (en) * | 2014-06-09 | 2015-12-09 | 北大方正集团有限公司 | Method and device for protecting image copyright |
CN106777177A (en) * | 2016-12-22 | 2017-05-31 | 百度在线网络技术(北京)有限公司 | Search method and device |
CN106709027A (en) * | 2016-12-28 | 2017-05-24 | 北京小米移动软件有限公司 | Picture recommending method and device |
Non-Patent Citations (2)
Title |
---|
唐红等: "面向未来互联网的语义感知服务匹配算法研究", 《重庆邮电大学学报(自然科学版)》 * |
杨佳等: "基于本体概念集合相似度的语义Web服务匹配", 《计算机技术与发展》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108363487A (en) * | 2018-01-29 | 2018-08-03 | 阿里巴巴集团控股有限公司 | Construction method, dreamland replay method and the device of dreamland playback model |
CN108363487B (en) * | 2018-01-29 | 2020-04-07 | 阿里巴巴集团控股有限公司 | Construction method of dream reproduction model, and dream reproduction method and device |
CN108596867A (en) * | 2018-05-09 | 2018-09-28 | 五邑大学 | A kind of picture bearing calibration and system based on ORB algorithms |
CN109783654A (en) * | 2018-12-11 | 2019-05-21 | 上海紫竹数字创意港有限公司 | A kind of media asset management system and method |
CN109885715A (en) * | 2019-01-25 | 2019-06-14 | 杭州凡闻科技有限公司 | Information processing method and system |
CN110910298A (en) * | 2019-10-23 | 2020-03-24 | 上海突进网络科技有限公司 | Picture processing method and system |
CN111462190A (en) * | 2020-04-20 | 2020-07-28 | 海信集团有限公司 | Intelligent refrigerator and food material input method |
CN111462190B (en) * | 2020-04-20 | 2023-11-17 | 海信集团有限公司 | Intelligent refrigerator and food material input method |
CN113747233A (en) * | 2021-08-24 | 2021-12-03 | 北京达佳互联信息技术有限公司 | Music replacement method and device, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN107545049B (en) | 2022-10-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107545049A (en) | Image processing method and related product | |
Pun et al. | A two-stage localization for copy-move forgery detection | |
Kadam et al. | [Retracted] Efficient Approach towards Detection and Identification of Copy Move and Image Splicing Forgeries Using Mask R‐CNN with MobileNet V1 | |
Chang et al. | Fabric defect detection based on pattern template correction | |
CN108062478A (en) | The malicious code sorting technique that global characteristics visualization is combined with local feature | |
Chu et al. | Image Retrieval Based on a Multi‐Integration Features Model | |
Türkyılmaz et al. | License plate recognition system using artificial neural networks | |
CN110502694A (en) | Lawyer's recommended method and relevant device based on big data analysis | |
CN107977948B (en) | Salient map fusion method facing community image | |
Kalyoncu et al. | GTCLC: leaf classification method using multiple descriptors | |
CN108460390A (en) | A kind of nude picture detection method of feature based study | |
Chugh et al. | [Retracted] Image Retrieval Using Different Distance Methods and Color Difference Histogram Descriptor for Human Healthcare | |
Unar et al. | New strategy for CBIR by combining low‐level visual features with a colour descriptor | |
Zhang et al. | Spectral-spatial classification of hyperspectral imagery based on deep convolutional network | |
Zhang et al. | The performance research of the data augmentation method for image classification | |
JP2009110526A (en) | Method and apparatus for analysing image | |
AU2021240205B1 (en) | Object sequence recognition method, network training method, apparatuses, device, and medium | |
Chang et al. | Deformed trademark retrieval based on 2D pseudo-hidden Markov model | |
Feng et al. | Image shadow detection and removal based on region matching of intelligent computing | |
Lu et al. | Detection of image seam carving using a novel pattern | |
Gurubelli et al. | Colour texture descriptor for CBIR of diseased tomato leaf images using modified local zigzag pattern | |
JP2009110525A (en) | Method and apparatus of searching for image | |
CN114913347A (en) | Clothes color identification method, system, equipment and storage medium | |
CN106156063B (en) | Correlation technique and device for object picture search results ranking | |
Zhu et al. | Image blind detection based on LBP residue classes and color regions |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |