CN106886783A - A kind of image search method and system based on provincial characteristics - Google Patents
A kind of image search method and system based on provincial characteristics Download PDFInfo
- Publication number
- CN106886783A CN106886783A CN201710048176.8A CN201710048176A CN106886783A CN 106886783 A CN106886783 A CN 106886783A CN 201710048176 A CN201710048176 A CN 201710048176A CN 106886783 A CN106886783 A CN 106886783A
- Authority
- CN
- China
- Prior art keywords
- image
- vision word
- sub
- retrieved
- regions
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Processing Or Creating Images (AREA)
Abstract
The present invention provides a kind of image search method and system based on provincial characteristics, and described method includes:S1, many sub-regions images are divided into by the image to be retrieved being input into;S2, extracts the provincial characteristics of each sub-district area image, and each provincial characteristics is quantified as into corresponding vision word;S3, travels through the corresponding vision word of each provincial characteristics, and number of each template image comprising the vision word is retrieved in database inverted list, using comprising the most template image of vision word number as retrieval result image.The present invention carries out subregion by image, extract the provincial characteristics of each area image, retrieval result is obtained by the matching of provincial characteristics, retrieved by local feature compared to existing, the provincial characteristics number of extraction greatly reduces compared to local feature number, the efficiency of image retrieval is improve, universality and autgmentability are more preferable.
Description
Technical field
The present invention relates to image retrieval technologies field, more particularly, to a kind of image retrieval side based on provincial characteristics
Method and system.
Background technology
Picture search based on content is one of key technology that big data epoch mass data information depth is utilized, and is also
The study hotspot of computer vision and MultiMedia Field, with important Research Significance and practical value, picture search technology exists
Also there is important application in real life.Security protection society is stepped into now, and camera has spread over each place in life, figure
It is that the police solve a case offer crucial clue as search can find specific target in the monitor video of magnanimity.Picture search
Technology also causes that daily life is more intelligent and convenient, and people can whenever and wherever possible shoot the clothes or article liked, then
Related commodity are searched in online shopping mall according to picture.
Most popular framework is the bag of words based on local invariant feature, the consistency of local feature at present for picture search
The complicated situations such as blocking in picture search, visual angle change can be well solved, is widely used in image recognition,
But local feature in some cases and is not being applied to.Local key point is generally detected at edge or corner point, and it can
Capture the details characteristic of the rigid objects in image, such as building, written pattern etc..Therefore, for rigid objects,
Bag of words based on local feature can reach preferable performance.If however, the content texture smoother in image, than
Such as sculpture and mollusk, then be only able to detect little key point in picture.Therefore, the object smooth for texture, base
Content in the local feature of key point can not express image well, it is local special when the retrieval of this kind of picture is processed
The poor-performing levied.Due to containing thousands of local features in each image, thus picture number it is more when, the efficiency of inquiry
It is slower.When the database of retrieval is very big, such as more than one hundred million picture, the memory cost of inverted list is very big, and this is also limited
The scalability of the bag of words based on local feature is made.
The content of the invention
The present invention provide it is a kind of overcome above mentioned problem or solve the above problems at least in part based on provincial characteristics
Image search method and system.
According to an aspect of the invention, there is provided a kind of image search method based on provincial characteristics, including:
S1, many sub-regions images are divided into by the image to be retrieved being input into;
S2, extracts the provincial characteristics of each sub-district area image, and each provincial characteristics is quantified as corresponding
Vision word;
S3, travels through the corresponding vision word of each provincial characteristics, and each Prototype drawing is retrieved in database inverted list
As the number comprising the vision word, using comprising the most template image of vision word number as retrieval result image.
According to another aspect of the present invention, a kind of image indexing system based on provincial characteristics is additionally provided, including:
Image division module, for the image to be retrieved being input into be divided into many sub-regions images;
Extraction quantization modules, the provincial characteristics for extracting each sub-district area image, and each region is special
Levy and be quantified as corresponding vision word;
Retrieval determining module, for traveling through the corresponding vision word of each provincial characteristics, examines in database inverted list
Number of each template image of rope comprising the vision word, will be defined as comprising the most template image of vision word number
Retrieval result image.
Beneficial effects of the present invention are:Subregion is carried out by image, the provincial characteristics of each area image is extracted, led to
The matching for crossing provincial characteristics obtains retrieval result, is retrieved by local feature compared to existing, the provincial characteristics number of extraction
Greatly reduce compared to local feature number, improve the efficiency of image retrieval, universality and autgmentability are more preferably, special relative to global
Levy, the retrieval rate of the method is higher, and effectiveness of retrieval is more or less the same with global characteristics, the present invention combines global characteristics
With the advantage of local feature, preferably retrieval performance is maintained in image retrieval procedure and the recall precision of original text is handed over.
Brief description of the drawings
Fig. 1 is the image search method flow chart based on provincial characteristics of one embodiment of the invention;
Fig. 2 is the image indexing system schematic diagram based on provincial characteristics of another embodiment of the present invention;
Fig. 3 is the specific block diagram representation of extraction quantization modules in another embodiment of the present invention;
Fig. 4 is the specific block diagram representation of the retrieval determining module in another embodiment of the present invention.
Specific embodiment
With reference to the accompanying drawings and examples, specific embodiment of the invention is described in further detail.Hereinafter implement
Example is not limited to the scope of the present invention for illustrating the present invention.
It is the image search method based on provincial characteristics of one embodiment of the invention referring to Fig. 1, including:
S1, many sub-regions images are divided into by the image to be retrieved being input into.
The image of retrieval to be checked is divided into many sub-regions images by the present embodiment first, specific to use super-pixel point
Image uniform to be retrieved is divided into many sub-regions images by the method cut.In actual applications, it is necessary to consider image retrieval
Accuracy rate and retrieval rate, an appropriate number of many sub-regions images, the sub-district after division are divided into by whole image to be retrieved
The quantity of area image is usually 20.
S2, extracts the provincial characteristics of each sub-regions image, and each provincial characteristics is quantified as into corresponding vision
Word.
After step S1 is divided to image to be retrieved, this step is directed to each sub-regions image, extracts each
The provincial characteristics of sub-district area image.The species of provincial characteristics is a lot, and the provincial characteristics in the present embodiment mainly includes deep learning
Feature and textural characteristics.Therefore, the present embodiment mainly extracts each when to each sub-regions image zooming-out provincial characteristics
The deep learning feature and textural characteristics of sub-district area image.The present embodiment extracts each using the method for Siamese network trainings
The deep learning feature of sub-regions image, and the deep learning feature that will be extracted generates deep learning feature code book, also extracts
The textural characteristics of each sub-regions image of image to be retrieved.
It is right by each deep learning characteristic quantification according to the deep learning feature of each sub-regions image for extracting
The vision word answered, wherein, deep learning feature is with the relation of vision word:One vision of multiple deep learning feature correspondences
Word.Wherein, the corresponding relation between deep learning feature and vision word is storable in database.It should be noted that
Before the retrieval of image is carried out, first by each template image in standard database according to same superpixel segmentation method
It is many sub-regions images to be evenly dividing, and extracts the deep learning spy of each sub-regions image of each template image
Seek peace textural characteristics, and each the deep learning characteristic quantification that will be extracted is corresponding vision word.It is according to vision word
The textural characteristics of vision word and corresponding template image mark and the sub-district area image are stored in data by index
In the inverted list in storehouse.
The present embodiment extracts the provincial characteristics of each sub-regions image of image to be retrieved, image to be retrieved each
Sub-district characteristic of field correspondence has a provincial characteristics, therefore, if image to be retrieved is divided into 20 sub-regions images, correspond to
The provincial characteristics of the image to be retrieved for extracting is 20, and compared to numerous local features for extracting image to be retrieved, extraction is treated
The number for retrieving the provincial characteristics of image can greatly reduce, therefore, the memory cost of database inverted list can also be greatly reduced.
S3, travels through the corresponding vision word of each provincial characteristics, each template image is retrieved in database inverted list and includes institute
The number of vision word is stated, using comprising the most template image of vision word number as retrieval result image.
Above-mentioned steps S2 extracts the deep learning feature and textural characteristics of each sub-regions image of image to be retrieved,
And by each deep learning characteristic quantification be corresponding vision word, obtain multiple visions corresponding with the image to be retrieved
Word.Retrieved in the inverted list of database according to each vision word, by the corresponding multiple visions of image to be retrieved
Word is traveled through one time, number of each template image in database of record inverted list comprising these vision words;And will
The corresponding textural characteristics of each sub-regions image of image to be retrieved are traveled through one time in the inverted list of database, record number
According to number of each template image in the inverted list of storehouse comprising these textural characteristics.Obtain each template image and include vision
The number of word and the number comprising textural characteristics, corresponding numerical value is calculated according to predetermined computation mode, numerical value is maximum
Template image is defined as retrieval result image.Specifically, the present embodiment is according to the weight coefficient of default vision word and default
Textural characteristics weight coefficient, the number of number and textural characteristics to vision word carries out weight calculation, obtains weight meter
Numerical value after calculation, the maximum template image of the numerical value that will be calculated is defined as final retrieval result image.
The image retrieval procedure of the present embodiment is illustrated with a specific example below:Assuming that by image to be retrieved according to
The method of super-pixel segmentation is divided into 20 sub-regions images, extracts the deep learning feature and texture of this 20 sub-regions image
Feature, obtains 20 deep learning features and 20 textural characteristics.Each deep learning feature be quantified as corresponding
Vision word, because multiple deep learning features may correspond to a vision word, therefore, 20 deep learning features may
15 vision words are quantified as, at this point for image to be retrieved, to that should have 15 vision words and 20 textural characteristics.It is directed to
15 vision words, each vision word is traveled through in the inverted list of database, it is assumed that in the inverted list of database
There are 100 template images, this 100 template images are numbered in inverted list, each template image there are multiple correspondences
Vision word and multiple textural characteristics.Each vision word is traveled through one time in inverted list, when all of vision word
After the completion of traversal, it is assumed that 8 vision words comprising image to be retrieved in template image 1, comprising retrieval image in template image 2
13 vision words, comprising 11 vision words etc. in template image 3, obtain each template in the inverted list of database
The number of the vision word comprising image to be retrieved in image.Similarly, by the line of each sub-regions image of image to be retrieved
Reason feature is traveled through in the inverted list of database, and each template image is comprising to be retrieved in obtaining the inverted list of database
The number of the textural characteristics of image.Vision list of each template image comprising image to be retrieved in database inverted list is obtained
The number of word and the number of textural characteristics, and weight coefficient and the weight of default textural characteristics according to default vision word
Coefficient, the number of number and textural characteristics to vision word carries out weight calculation, obtains the numerical value after weight calculation, will calculate
The maximum template image of numerical value out is defined as final retrieval result image.
In the present embodiment, the retrieval of image is carried out by provincial characteristics, due to the provincial characteristics of the image to be retrieved of extraction
Compare less, therefore, when carrying out image retrieval using provincial characteristics, retrieval rate comparing is carried out compared to using local feature
Hurry up, and provincial characteristics is easier to extract in the image of content texture smoother, is particularly suitable for content texture in image
The images match retrieval of smoother, just overcomes the shortcoming that local feature is not suitable for the smooth image retrieval of texture.Separately
Outward, the number of the provincial characteristics of the image to be retrieved that the present embodiment is extracted is multiple, therefore, examined compared to using global characteristics
Rope, the accuracy rate of retrieval can be higher, therefore, the method that the present embodiment is provided combines the advantage of global characteristics and local feature,
Preferably retrieval performance and retrieval rate higher is maintained in retrieving, compared to the image retrieval of local feature and complete
The image retrieval of office's feature, applicability is wider.
In addition, the present embodiment is extracted the deep learning feature and textural characteristics of image to be retrieved, by deep learning feature
Combine with textural characteristics, image is retrieved using two provincial characteristics, can more improve the accuracy rate of image retrieval.
It is the image indexing system based on provincial characteristics of another embodiment of the present invention, mainly including image referring to Fig. 2
Division module 21, extraction quantization modules 22, retrieval determining module 23 and memory module 24.
Image division module 21, for the image to be retrieved being input into be divided into many sub-regions images;
Extract quantization modules 22, the provincial characteristics for extracting each sub-regions image, and by each provincial characteristics
It is quantified as corresponding vision word;
Retrieval determining module 23, for traveling through the corresponding vision word of each provincial characteristics, in database inverted list
Number of each template image comprising the vision word is retrieved, will be determined comprising the most template image of vision word number
It is retrieval result image.
Wherein, image division module 21, divides image uniform to be retrieved specifically for the method according to super-pixel segmentation
It is many sub-regions images.
Referring to Fig. 3, extract quantization modules 22 and specifically include extraction unit 221 and quantifying unit 222, extraction unit 221 is used
In for each sub-regions image after being evenly dividing, the deep learning feature and texture for extracting each sub-regions image are special
Levy, obtain multiple deep learning features and multiple textural characteristics;
Quantifying unit 222, for being corresponding vision word by each deep learning characteristic quantification, obtains described to be checked
The corresponding vision word of each sub-regions image of rope image.
Described image division module 21, is additionally operable to:
Each template image in database is divided into many sub-regions images;
The extraction quantization modules 22, are additionally operable to:
Extract the deep learning feature and textural characteristics of each sub-regions image, and by each sub-regions image
Corresponding deep learning characteristic quantification is corresponding vision word;
Memory module 34, for being index according to vision word, by vision word and corresponding template image mark
Know and the textural characteristics of the sub-district area image are stored in the inverted list of database.
Referring to Fig. 4, retrieval determining module 23 specifically includes the first retrieval subelement 231, second and retrieves submodule 232, meter
Calculate unit 233 and determining unit 234;
First retrieval subelement 231, for the corresponding vision word of each sub-regions image for image to be retrieved,
Number of each template image comprising the vision word is retrieved from the inverted list of database;
Second retrieval subelement 232, for each the corresponding textural characteristics of block area image according to image to be retrieved,
Number of each template image comprising the textural characteristics is retrieved from database inverted list;
Computing unit 233, for the number of vision word that is included according to each template image and textural characteristics
Number, corresponding numerical value is calculated according to predetermined computation mode;
Determining unit 234, for the maximum template image of numerical value to be defined as into retrieval result image.
Wherein, computing unit has 233 bodies and is used for:
The weight coefficient of weight coefficient and default textural characteristics according to default vision word, to the vision word
Number and the number of the textural characteristics carry out weight calculation, obtain the numerical value after weight calculation.
A kind of image search method and system based on provincial characteristics that the present invention is provided, subregion is carried out by image,
The provincial characteristics of each area image is extracted, retrieval result is obtained by the matching of provincial characteristics, compared to existing by office
Portion's feature is retrieved, and the provincial characteristics number of extraction greatly reduces compared to local feature number, improve the efficiency of image retrieval,
More preferably, relative to global characteristics, the retrieval rate of the method is higher for universality and autgmentability, and effectiveness of retrieval is special with the overall situation
Levy and be more or less the same, the present invention combines the advantage of global characteristics and local feature, maintained in image retrieval procedure preferably
Retrieval performance and preferable recall precision;Meanwhile, the present invention is extracted the deep learning feature and textural characteristics of image to be retrieved,
Deep learning feature and textural characteristics are combined, image is retrieved using two provincial characteristics, can more improve image retrieval
Accuracy rate.
Finally, the present processes are only preferably embodiment, are not intended to limit the scope of the present invention.It is all
Within the spirit and principles in the present invention, any modification, equivalent substitution and improvements made etc. should be included in protection of the invention
Within the scope of.
Claims (12)
1. a kind of image search method based on provincial characteristics, it is characterised in that including:
S1, many sub-regions images are divided into by the image to be retrieved being input into;
S2, extracts the provincial characteristics of each sub-district area image, and each provincial characteristics is quantified as into corresponding vision
Word;
S3, travels through the corresponding vision word of each provincial characteristics, and each template image bag is retrieved in database inverted list
Number containing the vision word, using comprising the most template image of vision word number as retrieval result image.
2. the image search method of provincial characteristics is based on as claimed in claim 1, it is characterised in that the step S1 is specifically wrapped
Include:
Image uniform to be retrieved is divided into many sub-regions images by the method according to super-pixel segmentation.
3. the image search method of provincial characteristics is based on as claimed in claim 2, it is characterised in that the step S2 is specifically wrapped
Include:
S21, for each sub-regions image after being evenly dividing, extract each sub-regions image deep learning feature and
Textural characteristics, obtain multiple deep learning features and multiple textural characteristics;
S22, is corresponding vision word by each deep learning characteristic quantification, obtains each height of the image to be retrieved
The corresponding vision word of area image.
4. the image search method of provincial characteristics is based on as claimed in claim 3, it is characterised in that before the step S1 also
Including:
Each template image in database is divided into many sub-regions images, and extracts each sub-regions image
Deep learning feature and textural characteristics;
It is corresponding vision word by the corresponding deep learning characteristic quantification of each sub-regions image;
It is index according to vision word, by vision word and the line of corresponding template image mark and the sub-district area image
Reason feature is stored in the inverted list of database.
5. the image search method of provincial characteristics is based on as claimed in claim 4, it is characterised in that the step S3 is specifically wrapped
Include:
S31, for the corresponding vision word of each sub-regions image of image to be retrieved, retrieves from the inverted list of database
Number of each template image comprising the vision word;
S32, according to each corresponding textural characteristics of block area image of image to be retrieved, retrieves every from database inverted list
Number of one template image comprising the textural characteristics;
S33, the number of the vision word included according to each template image and the number of textural characteristics, according to predetermined computation
Mode calculates corresponding numerical value;
S34, retrieval result image is defined as by the maximum template image of numerical value.
6. the image search method of provincial characteristics is based on as claimed in claim 5, it is characterised in that the step S33 is specific
Including:
The weight coefficient of weight coefficient and default textural characteristics according to default vision word, to the individual of the vision word
The number of number and the textural characteristics carries out weight calculation, obtains the numerical value after weight calculation.
7. a kind of image indexing system based on provincial characteristics, it is characterised in that including:
Image division module, for the image to be retrieved being input into be divided into many sub-regions images;
Extract quantization modules, the provincial characteristics for extracting each sub-district area image, and by each provincial characteristics amount
Turn to corresponding vision word;
Retrieval determining module, for traveling through the corresponding vision word of each provincial characteristics, retrieves every in database inverted list
Number of one template image comprising the vision word, will be defined as retrieval comprising the most template image of vision word number
Result images.
8. the image indexing system of provincial characteristics is based on as claimed in claim 7, it is characterised in that described image divides mould
Block, specifically for:
Image uniform to be retrieved is divided into many sub-regions images by the method according to super-pixel segmentation.
9. the image indexing system of provincial characteristics is based on as claimed in claim 8, it is characterised in that the extraction quantization modules
Specifically include:
Extraction unit, for for each sub-regions image after being evenly dividing, extracting the depth of each sub-regions image
Learning characteristic and textural characteristics, obtain multiple deep learning features and multiple textural characteristics;
Quantifying unit, for being corresponding vision word by each deep learning characteristic quantification, obtains the image to be retrieved
The corresponding vision word of each sub-regions image.
10. the image indexing system of provincial characteristics is based on as claimed in claim 9, it is characterised in that described image divides mould
Block, is additionally operable to:
Each template image in database is divided into many sub-regions images;
The extraction quantization modules, are additionally operable to:
The deep learning feature and textural characteristics of each sub-regions image are extracted, and each sub-regions image is corresponding
Deep learning characteristic quantification be corresponding vision word;
Also include:
Memory module, vision word and corresponding template image for being index according to vision word, identified and be somebody's turn to do
The textural characteristics of sub-district area image are stored in the inverted list of database.
11. image indexing systems based on provincial characteristics as claimed in claim 10, it is characterised in that the retrieval determines mould
Block is specifically included:
First retrieval subelement, for the corresponding vision word of each sub-regions image for image to be retrieved, from data
Number of each template image comprising the vision word is retrieved in the inverted list in storehouse;
Second retrieval subelement, for each the corresponding textural characteristics of block area image according to image to be retrieved, from data
Number of each template image comprising the textural characteristics is retrieved in the inverted list of storehouse;
Computing unit, the number and the number of textural characteristics of the vision word for being included according to each template image, presses
Corresponding numerical value is calculated according to predetermined computation mode;
Determining unit, for the maximum template image of numerical value to be defined as into retrieval result image.
12. image indexing systems based on provincial characteristics as claimed in claim 11, it is characterised in that the computing unit tool
Body is used for:
The weight coefficient of weight coefficient and default textural characteristics according to default vision word, to the individual of the vision word
The number of number and the textural characteristics carries out weight calculation, obtains the numerical value after weight calculation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710048176.8A CN106886783B (en) | 2017-01-20 | 2017-01-20 | Image retrieval method and system based on regional characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710048176.8A CN106886783B (en) | 2017-01-20 | 2017-01-20 | Image retrieval method and system based on regional characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106886783A true CN106886783A (en) | 2017-06-23 |
CN106886783B CN106886783B (en) | 2020-11-10 |
Family
ID=59175958
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710048176.8A Active CN106886783B (en) | 2017-01-20 | 2017-01-20 | Image retrieval method and system based on regional characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106886783B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112488049A (en) * | 2020-12-16 | 2021-03-12 | 哈尔滨市科佳通用机电股份有限公司 | Fault identification method for foreign matter clamped between traction motor and shaft of motor train unit |
WO2021068524A1 (en) * | 2019-10-11 | 2021-04-15 | 平安科技(深圳)有限公司 | Image matching method and apparatus, computer device, and storage medium |
CN117874278A (en) * | 2024-03-11 | 2024-04-12 | 盛视科技股份有限公司 | Image retrieval method and system based on multi-region feature combination |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101344928A (en) * | 2007-07-12 | 2009-01-14 | 佳能株式会社 | Method and apparatus for confirming image area and classifying image |
US20130121600A1 (en) * | 2011-09-17 | 2013-05-16 | Zhe Lin | Methods and Apparatus for Visual Search |
CN104252625A (en) * | 2013-06-28 | 2014-12-31 | 河海大学 | Sample adaptive multi-feature weighted remote sensing image method |
US20160019440A1 (en) * | 2014-07-18 | 2016-01-21 | Adobe Systems Incorporated | Feature Interpolation |
CN105426533A (en) * | 2015-12-17 | 2016-03-23 | 电子科技大学 | Image retrieving method integrating spatial constraint information |
CN105447026A (en) * | 2014-08-27 | 2016-03-30 | 南京理工大学常熟研究院有限公司 | Web information extraction method based on minimum weight communication determining set in multi-view image |
CN105468596A (en) * | 2014-08-12 | 2016-04-06 | 腾讯科技(深圳)有限公司 | Image retrieval method and device |
CN105631471A (en) * | 2015-12-23 | 2016-06-01 | 西安电子科技大学 | Aurora sequence classification method with fusion of single frame feature and dynamic texture model |
CN105760507A (en) * | 2016-02-23 | 2016-07-13 | 复旦大学 | Cross-modal subject correlation modeling method based on deep learning |
CN106021250A (en) * | 2015-09-16 | 2016-10-12 | 展视网(北京)科技有限公司 | Image semantic information retrieval method based on keyword |
US20160335482A1 (en) * | 2014-11-07 | 2016-11-17 | Noblis, Inc. | Vector-based face recognition algorithm and image search system |
EP3115935A1 (en) * | 2015-06-29 | 2017-01-11 | Nokia Technologies Oy | A method, apparatus, computer program and system for image analysis |
-
2017
- 2017-01-20 CN CN201710048176.8A patent/CN106886783B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101344928A (en) * | 2007-07-12 | 2009-01-14 | 佳能株式会社 | Method and apparatus for confirming image area and classifying image |
US20130121600A1 (en) * | 2011-09-17 | 2013-05-16 | Zhe Lin | Methods and Apparatus for Visual Search |
CN104252625A (en) * | 2013-06-28 | 2014-12-31 | 河海大学 | Sample adaptive multi-feature weighted remote sensing image method |
US20160019440A1 (en) * | 2014-07-18 | 2016-01-21 | Adobe Systems Incorporated | Feature Interpolation |
CN105468596A (en) * | 2014-08-12 | 2016-04-06 | 腾讯科技(深圳)有限公司 | Image retrieval method and device |
CN105447026A (en) * | 2014-08-27 | 2016-03-30 | 南京理工大学常熟研究院有限公司 | Web information extraction method based on minimum weight communication determining set in multi-view image |
US20160335482A1 (en) * | 2014-11-07 | 2016-11-17 | Noblis, Inc. | Vector-based face recognition algorithm and image search system |
EP3115935A1 (en) * | 2015-06-29 | 2017-01-11 | Nokia Technologies Oy | A method, apparatus, computer program and system for image analysis |
CN106021250A (en) * | 2015-09-16 | 2016-10-12 | 展视网(北京)科技有限公司 | Image semantic information retrieval method based on keyword |
CN105426533A (en) * | 2015-12-17 | 2016-03-23 | 电子科技大学 | Image retrieving method integrating spatial constraint information |
CN105631471A (en) * | 2015-12-23 | 2016-06-01 | 西安电子科技大学 | Aurora sequence classification method with fusion of single frame feature and dynamic texture model |
CN105760507A (en) * | 2016-02-23 | 2016-07-13 | 复旦大学 | Cross-modal subject correlation modeling method based on deep learning |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021068524A1 (en) * | 2019-10-11 | 2021-04-15 | 平安科技(深圳)有限公司 | Image matching method and apparatus, computer device, and storage medium |
CN112488049A (en) * | 2020-12-16 | 2021-03-12 | 哈尔滨市科佳通用机电股份有限公司 | Fault identification method for foreign matter clamped between traction motor and shaft of motor train unit |
CN117874278A (en) * | 2024-03-11 | 2024-04-12 | 盛视科技股份有限公司 | Image retrieval method and system based on multi-region feature combination |
CN117874278B (en) * | 2024-03-11 | 2024-05-28 | 盛视科技股份有限公司 | Image retrieval method and system based on multi-region feature combination |
Also Published As
Publication number | Publication date |
---|---|
CN106886783B (en) | 2020-11-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107256262B (en) | Image retrieval method based on object detection | |
TWI528196B (en) | Similar image recognition method and apparatus | |
CN110413816B (en) | Color Sketch Image Search | |
CN102254015A (en) | Image retrieval method based on visual phrases | |
CN109101981B (en) | Loop detection method based on global image stripe code in streetscape scene | |
Srivastava et al. | Content-based image retrieval using moments of local ternary pattern | |
CN108182205A (en) | A kind of image method for quickly retrieving of the HASH algorithms based on SIFT | |
CN109919084B (en) | Pedestrian re-identification method based on depth multi-index hash | |
CN103929644A (en) | Video fingerprint database building method and device and video fingerprint recognition method and device | |
CN110334628B (en) | Outdoor monocular image depth estimation method based on structured random forest | |
CN113051984A (en) | Video copy detection method and apparatus, storage medium, and electronic apparatus | |
CN105404657A (en) | CEDD feature and PHOG feature based image retrieval method | |
Vimina et al. | A sub-block based image retrieval using modified integrated region matching | |
CN110956213A (en) | Method and device for generating remote sensing image feature library and method and device for retrieving remote sensing image | |
CN106886783A (en) | A kind of image search method and system based on provincial characteristics | |
EP1402403B1 (en) | Method of and apparatus for classifying an image | |
Kurchaniya et al. | Analysis of different similarity measures in image retrieval based on texture and shape | |
CN108694411B (en) | Method for identifying similar images | |
Liu et al. | Video copy detection by conducting fast searching of inverted files | |
Fierro-Radilla et al. | An effective visual descriptor based on color and shape features for image retrieval | |
Asif et al. | Composite description based on color vector quantization and visual primary features for CBIR tasks | |
CN104850600A (en) | Method and device for searching images containing faces | |
KR102054211B1 (en) | Method and system for video retrieval based on image queries | |
Cheng et al. | Visual object retrieval via block-based visual-pattern matching | |
KR100324255B1 (en) | Content based moving video retrieval method |
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 |