CN109960742A - The searching method and device of local message - Google Patents
The searching method and device of local message Download PDFInfo
- Publication number
- CN109960742A CN109960742A CN201910120165.5A CN201910120165A CN109960742A CN 109960742 A CN109960742 A CN 109960742A CN 201910120165 A CN201910120165 A CN 201910120165A CN 109960742 A CN109960742 A CN 109960742A
- Authority
- CN
- China
- Prior art keywords
- picture
- regional area
- checked
- segmentation result
- example segmentation
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to technical field of image processing, and in particular to the searching method and device of local message, wherein method includes: each local message for obtaining each picture in the picture and picture library to be checked with markup information;Picture to be checked is input to detection model, obtains position, the corresponding classification, example segmentation result in each regional area of each regional area in picture to be checked;Using the position of each regional area in markup information and picture to be checked, the classification of user's area-of-interest is determined;From picture to be checked and picture library, example segmentation result in the regional area corresponding to determining classification is extracted respectively;It based on the example segmentation result out of regional area that taken out in picture to be checked, is scanned in picture library, to obtain the search result of picture to be checked.The thought that example is divided is applied in local message search, can targetedly be searched for based on the interested region of user, can be improved the precision of picture search.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to the searching method and device of local message.
Background technique
It is an important project in computer vision field to scheme to search figure, main task is existed by retrieving single picture
The technology of similar image in image library provides the technology of associated picture retrieval for the mankind.It relates to computer vision, at image
Many technical fields such as reason, pattern-recognition and information processing, the either picture retrieval for mature face retrieval, network,
Or the license plate vehicle retrieval of monitoring field, which requires to put into a large amount of manpower, goes to handle.
In recent years, have extensively to scheme to search figure in fields such as intelligent video monitoring, Vehicular automatic driving, robot environment's perception
General application.For example, in public security big data system to scheme to search figure, be a vehicle image that user is provided as mesh
The image for marking vehicle searches the traveling record of target vehicle in the alert data record of bayonet or electricity of magnanimity.
For application angle, existing majority image retrieval related application is searched based on global characteristics, i.e., in sea
Image similar with target image is searched in the image data base of amount.However, should the image search method based on global characteristics
Local minutia can be ignored, and go to hold on the whole, so that it is relatively low to will lead to the image accuracy searched out, it is difficult to be
Safety monitoring provides quickly and has targetedly clue.
Summary of the invention
In view of this, being searched the embodiment of the invention provides a kind of searching method of local message and device with solving the overall situation
The low problem of searching accuracy caused by rope.
According in a first aspect, the embodiment of the invention provides a kind of searching methods of local message, comprising:
Obtain each local message of each picture in the picture and picture library to be checked with markup information, the office
Portion's information includes the position of regional area, corresponding classification, example segmentation result in regional area;Wherein, the markup information
For indicating the location information of user's area-of-interest;
The picture to be checked is input to the detection model, obtains each regional area in the picture to be checked
Position, corresponding classification, example segmentation result in each regional area;
Using the position of each regional area in the markup information and picture to be checked, user's region of interest is determined
The classification in domain;
Divide from each picture, being extracted respectively in the picture to be checked and the picture library corresponding to determining
Example segmentation result in the regional area of class;
It is every in the picture library based on the example segmentation result out of regional area that taken out in the picture to be checked
It is scanned in example segmentation result in the regional area taken out in one picture, to obtain the search of the picture to be checked
As a result.
The thought that example is divided is applied to local message and searched by the searching method of local message provided in an embodiment of the present invention
Suo Zhong is scanned in the local instance of each picture in picture library using the local instance of obtained picture to be checked,
This method can targetedly be searched for based on the interested region of user, can be improved the precision and effect of picture search
Rate.
With reference to first aspect, described to be based on taking from the picture to be checked in first aspect first embodiment
Example segmentation result in regional area out, example point in the regional area taken out in each picture in the picture library
It cuts in result and scans for, comprising:
Respectively to out of, regional area that taken out in each picture in the picture to be checked and the picture library
Example segmentation result carries out the extraction of example foreground features, to construct first partial provincial characteristics vector and several second parts
Provincial characteristics vector;Wherein, the first partial provincial characteristics vector is corresponding with the picture to be checked, second partial zones
Characteristic of field vector is corresponding with each picture in the picture library;
Based on the first partial provincial characteristics vector and each second local features vector, calculate similar
Degree;
Calculated result based on the similarity extracts corresponding picture from the picture library.
The searching method of local message provided in an embodiment of the present invention is carried out by example segmentation result in localized region
The extraction of example foreground features reduces influence of the background to local search, improves the accuracy of search.
First embodiment with reference to first aspect, in first aspect second embodiment, it is described respectively to from it is described to
Before example segmentation result carries out example in the regional area taken out in each picture in inquiry picture and the picture library
The extraction of scape feature, to construct first partial provincial characteristics vector and several second local features vectors, comprising:
Pixel in first example segmentation result and the second example segmentation result is divided into the pixel value in the region of background
Zero setting;Wherein, the first example segmentation result is the example segmentation result out of regional area that take out in the picture to be checked,
Second example segmentation result is the example segmentation result out of regional area that take out in picture each in the picture library;
To after zero setting the first example segmentation result and the second example segmentation result carry out pond, it is identical to obtain size
The first partial provincial characteristics vector and the second local features vector.
It is special to obtain the identical regional area of size by pond for the searching method of local message provided in an embodiment of the present invention
Vector is levied, i.e., floating type image pixel value is obtained by pond, the search for subsequent local message provides condition.
First embodiment with reference to first aspect calculates institute using following formula in first aspect third embodiment
State similarity:
Wherein, Feat1 is first partial provincial characteristics vector, and Feat2 is the second local features vector.
With reference to first aspect, described using the markup information and to be checked in the 4th embodiment of first aspect
The position of each regional area in picture determines the classification of user's area-of-interest, comprising:
Calculate the friendship of each regional area and ratio in user's area-of-interest and the picture to be checked;
Friendship and ratio based on each regional area in user's area-of-interest and the picture to be checked, determine the use
The position of the corresponding regional area of family area-of-interest;
The classification for corresponding to the position for the regional area determined is extracted, to obtain point of user's area-of-interest
Class.
The searching method of local message provided in an embodiment of the present invention, by user's area-of-interest and picture to be checked
The friendship of each regional area and ratio determine the position for being used for the corresponding regional area of area-of-interest, due to handing over and than can be very big
Reflect to degree area accounting of the regional area in user's area-of-interest, therefore using handing over and interested than improving user
The accuracy that the corresponding regional area position in region determines.
With reference to first aspect, in the 5th embodiment of first aspect, the training process of the detection model includes following
Step:
Initialize the parameter of the neural network;
The sample image is input to the neural network, position, the correspondence of regional area are exported by propagated forward
Classification;
Example segmentation is carried out to the regional area using mask branch;
Position based on the regional area, corresponding classification and example segmentation as a result, with the sample image
Mark value is compared, and optimizes the parameter of the neural network.
The searching method of local message provided in an embodiment of the present invention reduces the dry of ambient noise using mask branch
It disturbs, provides advantageous clue for the search of regional area.
5th embodiment with reference to first aspect, in first aspect sixth embodiment, the loss of the neural network
Function is the damage of the loss function of the regional area position, the loss function of the corresponding classification and example segmentation
Lose the sum of function.
According to second aspect, the embodiment of the invention provides a kind of searchers of local message, comprising:
First obtains module, obtains each office of each picture in the picture and picture library to be checked with markup information
Portion's information, the local message include the position of regional area, corresponding classification, example segmentation result in regional area;Wherein,
The markup information is used to indicate the location information of user's area-of-interest;
Input module obtains in the picture to be checked for the picture to be checked to be input to the detection model
The position of each regional area, corresponding classification, example segmentation result in each regional area;
Determining module determines institute for the position using each regional area in the markup information and picture to be checked
State the classification of user's area-of-interest;
Extraction module, for from each picture, being extracted respectively in the picture to be checked and the picture library
Corresponding to example segmentation result in the regional area of determining classification;
Search module, for based on the example segmentation result out of regional area that taken out in the picture to be checked,
Scanned in example segmentation result in the regional area taken out in each picture in the picture library, with obtain it is described to
Inquire the search result of picture.
The thought that example is divided is applied to local message and searched by the searcher of local message provided in an embodiment of the present invention
Suo Zhong is scanned in the local instance of each picture in picture library using the local instance of obtained picture to be checked,
The device can targetedly be searched for based on the interested region of user, can be improved the precision and effect of picture search
Rate.
According to the third aspect, the embodiment of the invention provides a kind of electronic equipment, comprising: memory and processor, it is described
Connection is communicated with each other between memory and the processor, computer instruction is stored in the memory, and the processor is logical
It crosses and executes the computer instruction, thereby executing office described in any one of first aspect or first aspect embodiment
The searching method of portion's information.
It is described computer-readable the embodiment of the invention provides a kind of computer readable storage medium according to fourth aspect
Storage medium stores computer instruction, and the computer instruction is for making the computer execute first aspect or first aspect
Any one embodiment described in local message searching method.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 is the flow chart of the searching method of local message according to an embodiment of the present invention;
Fig. 2 is the flow chart of the searching method of local message according to an embodiment of the present invention;
Fig. 3 is the flow chart of the searching method of local message according to an embodiment of the present invention;
Fig. 4 is the flow chart of the training method of detection model according to an embodiment of the present invention;
Fig. 5 is the block diagram of the training method of detection model according to an embodiment of the present invention;
Fig. 6 is the block diagram of the searching method of local message according to an embodiment of the present invention;
Fig. 7 is the method flow diagram of training stage and search phase according to an embodiment of the present invention;
Fig. 8 is the search result schematic diagram of local message according to an embodiment of the present invention;
Fig. 9 is the structural block diagram of the searcher of local message according to an embodiment of the present invention;
Figure 10 is the hardware structural diagram of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those skilled in the art are not having
Every other embodiment obtained under the premise of creative work is made, shall fall within the protection scope of the present invention.
According to embodiments of the present invention, a kind of searching method embodiment of local message is provided, it should be noted that attached
The step of process of figure illustrates can execute in a computer system such as a set of computer executable instructions, though also,
So logical order is shown in flow charts, but in some cases, it can be to be different from shown by sequence execution herein
Or the step of description.
Provide a kind of searching method of local message in the present embodiment, can be used for above-mentioned electronic equipment, as mobile phone,
Tablet computer etc., Fig. 1 is the flow chart of the searching method of local message according to an embodiment of the present invention, as shown in Figure 1, the process
Include the following steps:
S11 obtains the local message of each picture in the picture and picture library to be checked with markup information.
Wherein, the markup information is used to indicate the location information of user's area-of-interest.For example, picture to be checked is behaved
Member's picture, user's area-of-interest are the shoes with Nike mark.So, user needs before the search for carrying out local message
The shoes with Nike mark are marked out on personnel's picture, using the markup information as personnel's picture.It, can specifically in mark
It is manually marked with treating inquiry picture using characteristics of image label select tool (for example, LabelImg), markup information is
(Xp,Yp,Lp,Wp), wherein (Xp,Yp) it is characterized the coordinate information in the upper left corner, (Lp,Wp) be the length of pixel that this feature accounts for and
It is wide;It can also carry out manual mark etc..
It further, include several pictures in picture library, each picture in picture library accessed by electronic equipment
Local message is that each picture is input to what detection model obtained in picture library, and the detection model is to be instructed using neural network
Practice what several sample images obtained.
Specifically, the input of the detection model is picture, exports the position for regional area each in picture, corresponding point
Example segmentation result in class, regional area.Wherein, corresponding classification is set when being trained to detection model in advance,
It can set to preset and be classified as K class.For example, input one personnel's picture, the regional area of output include four limbs, head, the upper part of the body,
Lower part of the body etc..Detection model selects each regional area using recurrence frame (rectangle frame) frame in output, then represents each
Example segmentation result on personnel's image (that is, if exist in classification described in regional area, and each regional area of output
Multiple people, then can individually be partitioned into everyone).
It should be noted that the local message of each picture can be and be previously entered detection model and obtain in picture library
, it is also possible to input what detection model obtained when needing the search for carrying out local message.
Picture to be checked is input to detection model by S12, obtains the position, right of each regional area in picture to be checked
Example segmentation result in the classification answered, each regional area.
Electronic equipment is input to detection after getting the picture to be checked with markup information, by the picture to be checked
The local message of picture to be checked can be obtained in model, position, corresponding classification and each partial zones including each regional area
Example segmentation result in domain.
S13 determines user's area-of-interest using the position of each regional area in markup information and picture to be checked
Classification.
Markup information due to being input to detection model is intended to indicate that the interested location information of user, while to be checked
Model can obtain the position of each regional area to picture later after testing, then passing through position represented by comparison markup information
The position of information and each regional area can determine that it is which regional area corresponding to markup information, so that it is determined that bid
Infuse the classification (that is, the described classification for determining user's area-of-interest) of the corresponding regional area of information.
Wherein it is possible to by the central point of location information represented by correlation marker information, the position with each regional area
Central point, determine the corresponding regional area of markup information;It is also possible to by successively seeking position represented by flag information
Friendship and ratio of the information with the position of each regional area, determine the corresponding regional area of markup information.
Electronic equipment in determining picture to be checked after the position of the corresponding regional area of user's area-of-interest, by
It is corresponding in the position of regional area and the classification of regional area, then utilizing the position of determined regional area can determine
The classification of regional area out.Therefore, electronic equipment can determine the corresponding classification of user's area-of-interest in picture to be checked.
S14 is extracted respectively from picture each in picture to be checked and picture library corresponding to determining classification
Example segmentation result in regional area.
Electronic equipment, can be to be checked in determining picture to be checked after classification belonging to user's area-of-interest
Extract example segmentation result in the regional area corresponding to the classification in picture in each regional area in example segmentation result;
In the same way, electronic equipment can extract the regional area corresponding to the classification from each picture of picture library
Interior example segmentation result.
S15, based on the example segmentation result out of regional area that taken out in picture to be checked, each in picture library
It is scanned in example segmentation result in the regional area taken out in picture, to obtain the search result of picture to be checked.
Electronic equipment example segmentation result in the obtained regional area corresponding to user's area-of-interest in S14
On the basis of, search example segmentation result out of regional area that take out in picture each in picture library obtains figure to be checked
The search result of piece.Wherein it is possible to be example in the regional area for indicate in the form of feature vector user's area-of-interest
Example segmentation result in the regional area taken out in each picture in segmentation result and picture library, it is then special by calculating
The similarity between vector is levied, extracts the search result of picture to be checked from picture library using similarity;It is first right to be also possible to
The partial zones taken out in each picture in example segmentation result and picture library in the regional area of user's area-of-interest
Example segmentation result carries out foreground features extraction in domain, then indicates using feature vector, calculates similarity;Or it can also be with
The search etc. of picture to be checked is carried out using other modes.
The thought that example is divided is applied to local message search by the searching method of local message provided in this embodiment
In, using the local instance of obtained picture to be checked, scanned in the local instance of each picture in picture library, it should
Method can targetedly be searched for based on the interested region of user, can be improved the precision and effect of picture search
Rate.
A kind of searching method of local message is additionally provided in the present embodiment, can be used for above-mentioned electronic equipment, such as hand
Machine, tablet computer etc., Fig. 1 are the flow charts of the searching method of local message according to an embodiment of the present invention, as shown in Fig. 2, should
Process includes the following steps:
S21 obtains the local message of each picture in the picture and picture library to be checked with markup information.
The local message includes the position of each regional area, corresponding classification, example segmentation result in regional area;Its
In, the markup information is used to indicate the location information of user's area-of-interest;The local message is every in the picture library
One picture is input to what detection model obtained;The detection model is to be obtained using several sample images of neural metwork training
's.
The S11 of embodiment illustrated in fig. 1 is referred to, details are not described herein.
Picture to be checked is input to detection model by S22, obtains the position, right of each regional area in picture to be checked
Example segmentation result in the classification answered, each regional area.
The S12 of embodiment illustrated in fig. 1 is referred to, details are not described herein.
S23 determines user's area-of-interest using the position of each regional area in markup information and picture to be checked
Classification.
The S13 of embodiment illustrated in fig. 1 is referred to, details are not described herein.
S24 is extracted respectively from picture each in picture to be checked and picture library corresponding to determining classification
Example segmentation result in regional area.
The S14 of embodiment illustrated in fig. 1 is referred to, details are not described herein.
S25, based on the example segmentation result out of regional area that taken out in picture to be checked, each in picture library
It is scanned in example segmentation result in the regional area taken out in picture, to obtain the search result of picture to be checked.
Electronic equipment calculates the form of similarity using feature vector, treats inquiry picture and scans for.Specifically, it wraps
It includes:
S251, respectively to the example out of regional area that taken out in picture each in picture to be checked and picture library
Segmentation result carries out the extraction of example foreground features, to construct first partial provincial characteristics vector and several second regional areas
Feature vector.
Wherein, first partial provincial characteristics vector is corresponding with picture to be checked, the second local features vector and picture
Each picture is corresponding in library.
Picture is input to after detection model, not only exports the position of each regional area of picture, corresponding classification and each
Example segmentation result in regional area, additionally it is possible to obtain the global characteristics corresponding to picture.Electronic equipment utilizes the overall situation of picture
Feature can determine which pixel is to belong to background, which pixel is to belong to prospect.
Therefore, in the regional area that electronic equipment takes out in each picture in treating inquiry picture and picture library
When example segmentation result carries out the extraction of example foreground features, it may include steps of:
(1) pixel in the first example segmentation result and the second example segmentation result is divided into the pixel in the region of background
It is worth zero setting.Wherein, the first example segmentation result is the example segmentation result out of regional area that take out in picture to be checked, the
Two example segmentation results are the example segmentation result out of regional area that take out in picture each in picture library.
(2) to after zero setting the first example segmentation result and the second example segmentation result carry out pond, to obtain size
Identical first partial provincial characteristics vector and the second local features vector.
Wherein it is possible to carry out bilinear interpolation by the way of pool area (ROI Align Pooling), size is obtained
Identical first partial provincial characteristics vector and the second local features vector.Specifically, first partial provincial characteristics to
Amount corresponds to picture to be checked;The quantity of second local features vector is identical as the quantity of picture in picture library, i.e. picture
Each picture corresponds to a second local features vector in library.
S252 calculates similarity based on first partial provincial characteristics vector and each second local features vector.
Wherein it is possible to be to calculate the distance between two feature vectors, similarity is indicated using distance;It is also possible to use
Following formula calculates similarity:
Wherein, Similarity is similarity, and Feat1 is first partial provincial characteristics vector, and Feat2 is the second partial zones
Characteristic of field vector.
When calculating similarity using above-mentioned formula, need successively to calculate first partial provincial characteristics vector and each the
Similarity between two local features vectors, to obtain similarity identical with picture number in picture library.
S253 extracts corresponding picture based on the calculated result of similarity from picture library.
Electronic equipment can be ranked up similarity after obtaining several similarities, and it is highest more to choose similarity
Search result of a picture as picture to be checked.
Compared with embodiment illustrated in fig. 1, the searching method of local message provided in this embodiment, by localized region
Example segmentation result carries out the extraction of example foreground features, reduces influence of the background to local search, improves the standard of search
True property.
A kind of searching method of local message is additionally provided in the present embodiment, can be used for above-mentioned electronic equipment, such as hand
Machine, tablet computer etc., Fig. 1 are the flow charts of the searching method of local message according to an embodiment of the present invention, as shown in figure 3, should
Process includes the following steps:
S31 obtains the local message of each picture in the picture and picture library to be checked with markup information.
The local message includes the position of each regional area, corresponding classification, example segmentation result in regional area;Its
In, the markup information is used to indicate the location information of user's area-of-interest;The local message is every in the picture library
One picture is input to what detection model obtained;The detection model is to be obtained using several sample images of neural metwork training
's.
The S21 of embodiment illustrated in fig. 2 is referred to, details are not described herein.
Picture to be checked is input to detection model by S32, obtains the position, right of each regional area in picture to be checked
Example segmentation result in the classification answered, each regional area.
The S22 of embodiment illustrated in fig. 2 is referred to, details are not described herein.
S33 determines user's area-of-interest using the position of each regional area in markup information and picture to be checked
Classification.
Electronic equipment determines the classification of user's area-of-interest by the way of friendship and ratio.Specifically, comprising:
S331 calculates the friendship of the position of each regional area and ratio in user's area-of-interest and picture to be checked.
Since markup information is (Xp,Yp,Lp,Wp), the interested region of user is represented using the markup information;Each office
The location information in portion region is exported from detection model, and each regional area can be represented.Electronic equipment is by calculating user
In area-of-interest and picture to be checked the friendship of the position of each regional area and than (that is, Intersection-over-Union,
Referred to as IOU), with the position for the corresponding regional area of subsequent determining user's area-of-interest.
S332, friendship and ratio based on each regional area in user's area-of-interest and the picture to be checked, determines user
The position of the corresponding regional area of area-of-interest.
Corresponding to each regional area, successively calculates and hand over and compare, to determine that user is interested using calculated size
The position of the corresponding regional area in region.
Alternatively, optionally it is determined that can also feel in conjunction with user when the position of the corresponding regional area of user's area-of-interest
The central point of the position of the central point in interest region and each regional area.For example, a part of office can be excluded first with central point
Then portion region recycles the position of above-mentioned friendship and regional area more corresponding than determining user's area-of-interest.
S333 extracts the classification for corresponding to the position for the regional area determined, to obtain point of user's area-of-interest
Class.
Electronic equipment utilizes regional area after the position for determining the corresponding regional area of user's area-of-interest
Corresponding relationship between the corresponding classification in position, that is, can determine that the classification of the position of regional area, so as to obtain
The classification of user's area-of-interest.
S34 is extracted respectively from picture each in picture to be checked and picture library corresponding to determining classification
Example segmentation result in regional area.
The S24 of embodiment illustrated in fig. 2 is referred to, details are not described herein.
S35, based on the example segmentation result out of regional area that taken out in picture to be checked, each in picture library
It is scanned in example segmentation result in the regional area taken out in picture, to obtain the search result of picture to be checked.
The S25 of embodiment illustrated in fig. 2 is referred to, details are not described herein.
Compared with embodiment illustrated in fig. 2, the searching method of local message provided in this embodiment passes through user's region of interest
The friendship of each regional area and ratio in domain and picture to be checked, determine the position for being used for the corresponding regional area of area-of-interest, by
In handing over and than can dramatically reflect area accounting of the regional area in user's area-of-interest, therefore using handing over and compare
Improve the accuracy that the corresponding regional area position of user's area-of-interest determines.
As a kind of optional embodiment of the present embodiment, as shown in figure 4, the training process of the detection model include with
Lower step:
S41 initializes the parameter of neural network.
Constructed neural network out can be MASK RCNN, initialize to parameters in the network, and setting is just
Initial value;The default of the network, which is arranged, is classified as K class simultaneously.Wherein, constructed neural network out is as shown in Figure 5.
Sample image is input to neural network by S42, passes through the position of propagated forward output regional area, corresponding point
Class.
Several sample images constitute data set, and data set is divided into training set, verifying collection, test set, labeled data collection
And data augmentation processing is carried out, including flip horizontal, rotation plus make an uproar, translate, luminance contrast adjustment etc., and guarantee each figure
The input scaled of picture subtracts averaging operation to 224 × 224 pixels, to what image was unified.
Specifically, firstly, using Resnet50 as basic convolutional neural networks (Convolutional Neural
Network, referred to as CNN), extract the global depth feature of input picture.Secondly, similar Faster RCNN, is tied using RPN
Structure generates anchor (anchors), generates the suggestion areas (Region Proposals) of target.Again, in conjunction with depth global characteristics,
Further progress ROI Align Pooling operation, generates an equal amount of pond result characteristic pattern to each suggestion areas.
Finally, passing through the full convolutional network (Fully of design based on the fixed size feature that ROI Align Pooling is obtained
Convolutional Networks, FCN) generate position, classification that specific regional area is classified.
S43 carries out example segmentation using mask branch localized region.
The detection model is based on mask branch (Mask branch) and generates pixel segmentation result to get local to corresponding to
The example segmentation result in region.
S44, position based on regional area, corresponding classification and example segmentation as a result, mark with sample image
Value is compared, the parameter of optimization neural network.
The result of position, corresponding classification and example segmentation based on regional area mentions compared with true mark value
For back-propagation algorithm training deep learning network.There are three loss functions, including regional area Classification Loss function altogether for network
Lcls, regional area position loss function LboxAnd the loss function L of example segmentationmask.Loss function be above-mentioned three it
With expression are as follows:
L=Lcls+Lbox+Lmask。
The searching method of local message provided in an embodiment of the present invention reduces the dry of ambient noise using mask branch
It disturbs, provides advantageous clue for the search of regional area.
As a kind of specific embodiment of the present embodiment, for personnel's local message, as shown in fig. 7, left side in Fig. 7
For the flow chart of detection model training stage, the right side Fig. 7 is the flow chart of search phase, the searching method of the local message
It is described in detail below:
The output of step (1) local region information: each office is exported based on the detection network trained for the picture of input
Portion presorts in region, its corresponding position, example segmentation result in regional area.
Step (2) classification foreground features to be checked are extracted: for image to be checked, according to the area-of-interest of user query with
The coordinate frame position (position of each regional area) of all default classification determines which kind of corresponding default classification.
The corresponding image to be checked of the classification and picture library image are carried out the feature extraction of example prospect by step (3).
Specifically, the region for the pixel for corresponding to the corresponding regional area of user's area-of-interest being divided into background is whole
Zero setting, the feature after obtaining background zero setting search for part as the local features vector for corresponding to image to be checked, such as Fig. 6
Shown in schematic diagram.
Step (4) similarity calculation: it is based on global characteristics and regional area relevant information, takes local feature.Part is special
Sign, which is extracted, obtains floating type image pixel value using bilinear interpolation, with reference to ROI Align Pooling.Successively matching primitives are looked into
Ask the cosine similarity of image and each picture library image:
Step (5) local message detects search result output: according to the similarity for extracting feature with query image, by picture
Picture descending arrangement in library is as output.
A kind of demonstration effect of the searching method of personnel's local message of the present invention is as shown in Figure 8.As seen from the figure, according to left side
Inquiring picture, (wherein, the corresponding local message of the first row is the tail portion of electric bicycle, and the corresponding local message of the second row is row
The jacket of people, the corresponding local message of the third line are the lower part of the body of pedestrian), right side gives the most similar in 10 picture libraries
Picture gets a desired effect.
Additionally provide a kind of searcher of local message in the present embodiment, the device for realizing above-described embodiment and
Preferred embodiment, the descriptions that have already been made will not be repeated.As used below, predetermined function may be implemented in term " module "
The combination of the software and/or hardware of energy.It is hard although device described in following embodiment is preferably realized with software
The realization of the combination of part or software and hardware is also that may and be contemplated.
The present embodiment provides a kind of searchers of local message, as shown in Figure 9, comprising:
First obtain module 61, obtain have markup information picture and picture library to be checked in each picture it is each
Local message, the local message include the position of regional area, corresponding classification, example segmentation result in regional area;Its
In, the markup information is used to indicate the location information of user's area-of-interest;The local message is every in the picture library
One picture is input to what detection model obtained;The detection model is to be obtained using several sample images of neural metwork training
's.
Input module 62 obtains the picture to be checked for the picture to be checked to be input to the detection model
In the position of each regional area, corresponding classification, example segmentation result in each regional area.
Determining module 63 is determined for the position using each regional area in the markup information and picture to be checked
The classification of user's area-of-interest.
Extraction module 64, for from each picture, being extracted respectively in the picture to be checked and the picture library
Correspond to example segmentation result in the regional area of determining classification out.
Search module 65, for based on the example segmentation result out of regional area that taken out in the picture to be checked,
It is scanned in example segmentation result in the regional area taken out in each picture in the picture library, it is described to obtain
The search result of picture to be checked.
The thought that example is divided is applied to local message search by the searcher of local message provided in this embodiment
In, using the local instance of obtained picture to be checked, scanned in the local instance of each picture in picture library, it should
Device can targetedly be searched for based on the interested region of user, can be improved the precision and effect of picture search
Rate.
The searcher of local message in the present embodiment is presented in the form of functional unit, and unit here refers to
ASIC circuit, execute one or more softwares or fixed routine processor and memory and/or other above-mentioned function can be provided
The device of energy.
The further function description of above-mentioned modules is identical as above-mentioned corresponding embodiment, and details are not described herein.
The embodiment of the present invention also provides a kind of electronic equipment, the searcher with above-mentioned local message shown in Fig. 9.
Referring to Fig. 10, Figure 10 is the structural schematic diagram for a kind of electronic equipment that alternative embodiment of the present invention provides, such as scheme
Shown in 10, which may include: at least one processor 71, such as CPU (Central Processing Unit, in
Central processor), at least one communication interface 73, memory 74, at least one communication bus 72.Wherein, communication bus 72 is used for
Realize the connection communication between these components.Wherein, communication interface 73 may include display screen (Display), keyboard
(Keyboard), optional communication interface 73 can also include standard wireline interface and wireless interface.Memory 74 can be high speed
RAM memory (Random Access Memory, effumability random access memory), is also possible to non-labile storage
Device (non-volatile memory), for example, at least a magnetic disk storage.Memory 74 optionally can also be at least one
It is located remotely from the storage device of aforementioned processor 71.Wherein processor 71 can be with device described in conjunction with Figure 9, in memory 74
Application program is stored, and processor 71 calls the program code stored in memory 74, for executing any of the above-described method step
Suddenly.
Wherein, communication bus 72 can be Peripheral Component Interconnect standard (peripheral component
Interconnect, abbreviation PCI) bus or expanding the industrial standard structure (extended industry standard
Architecture, abbreviation EISA) bus etc..Communication bus 72 can be divided into address bus, data/address bus, control bus etc..
Only to be indicated with a thick line in Figure 10, it is not intended that an only bus or a type of bus convenient for indicating.
Wherein, memory 74 may include volatile memory (English: volatile memory), such as arbitrary access
Memory (English: random-access memory, abbreviation: RAM);Memory also may include nonvolatile memory (English
Text: non-volatile memory), for example, flash memory (English: flash memory), hard disk (English: hard disk
Drive, abbreviation: HDD) or solid state hard disk (English: solid-state drive, abbreviation: SSD);Memory 74 can also include
The combination of the memory of mentioned kind.
Wherein, processor 71 can be central processing unit (English: central processing unit, abbreviation: CPU),
The combination of network processing unit (English: network processor, abbreviation: NP) or CPU and NP.
Wherein, processor 71 can further include hardware chip.Above-mentioned hardware chip can be specific integrated circuit
(English: application-specific integrated circuit, abbreviation: ASIC), programmable logic device (English:
Programmable logic device, abbreviation: PLD) or combinations thereof.Above-mentioned PLD can be Complex Programmable Logic Devices
(English: complex programmable logic device, abbreviation: CPLD), field programmable gate array (English:
Field-programmable gate array, abbreviation: FPGA), Universal Array Logic (English: generic array
Logic, abbreviation: GAL) or any combination thereof.
Optionally, memory 74 is also used to store program instruction.Processor 71 can be instructed with caller, realize such as this Shen
Please local message shown in Fig. 1 to 4 embodiments searching method.
The embodiment of the invention also provides a kind of non-transient computer storage medium, the computer storage medium is stored with
The search of the local message in above-mentioned any means embodiment can be performed in computer executable instructions, the computer executable instructions
Method.Wherein, the storage medium can be magnetic disk, CD, read-only memory (Read-Only Memory, ROM), random
Storage memory (Random Access Memory, RAM), flash memory (Flash Memory), hard disk (Hard Disk
Drive, abbreviation: HDD) or solid state hard disk (Solid-State Drive, SSD) etc.;The storage medium can also include above-mentioned
The combination of the memory of type.
Although being described in conjunction with the accompanying the embodiment of the present invention, those skilled in the art can not depart from the present invention
Spirit and scope in the case where make various modifications and variations, such modifications and variations are each fallen within by appended claims institute
Within the scope of restriction.
Claims (10)
1. a kind of searching method of local message characterized by comprising
Obtain the local message of each picture in the picture and picture library to be checked with markup information, the local message
Example segmentation result in position, corresponding classification, regional area including each regional area;Wherein, the markup information is used for
Indicate the location information of user's area-of-interest;
The picture to be checked is input to the detection model that training obtains in advance, obtains each part in the picture to be checked
The position in region, corresponding classification, example segmentation result in each regional area;
Using the position of each regional area in the markup information and picture to be checked, user's area-of-interest is determined
Classification;
From the classification in each picture, taken out respectively in the picture to be checked and the picture library corresponding to determination
Example segmentation result in regional area;
Based on the example segmentation result out of regional area that taken out in the picture to be checked, each in the picture library
It is scanned in example segmentation result in the regional area taken out in picture, to obtain the search knot of the picture to be checked
Fruit.
2. the method according to claim 1, wherein described based on the office taken out from the picture to be checked
Example segmentation result in portion region, example segmentation result in the regional area taken out in each picture in the picture library
In scan for, comprising:
Respectively to the example out of regional area that taken out in each picture in the picture to be checked and the picture library
Segmentation result carries out the extraction of example foreground features, to construct first partial provincial characteristics vector and several second regional areas
Feature vector;Wherein, the first partial provincial characteristics vector is corresponding with the picture to be checked, and second regional area is special
It is corresponding with each picture in the picture library to levy vector;
Based on the first partial provincial characteristics vector and each second local features vector, similarity is calculated;
Calculated result based on the similarity extracts corresponding picture from the picture library.
3. according to the method described in claim 2, it is characterized in that, described respectively to from the picture to be checked and the figure
Example segmentation result carries out the extraction of example foreground features in the regional area taken out in each picture in valut, with building
First partial provincial characteristics vector and several second local features vectors, comprising:
Pixel in first example segmentation result and the second example segmentation result is divided into the pixel value zero setting in the region of background;
Wherein, the first example segmentation result be out of regional area that taken out in the picture to be checked example segmentation result, second
Example segmentation result is the example segmentation result out of regional area that take out in picture each in the picture library;
To after zero setting the first example segmentation result and the second example segmentation result carry out pond, to obtain the identical institute of size
State first partial provincial characteristics vector and the second local features vector.
4. according to the method described in claim 2, it is characterized in that, calculating the similarity using following formula:
Wherein, Feat1 is first partial provincial characteristics vector, and Feat2 is the second local features vector.
5. the method according to claim 1, wherein described using in the markup information and picture to be checked
The position of each regional area determines the classification of user's area-of-interest, comprising:
Calculate the friendship of each regional area and ratio in user's area-of-interest and the picture to be checked;
Friendship and ratio based on each regional area in user's area-of-interest and the picture to be checked determine user's sense
The position of the corresponding regional area in interest region;
The classification for corresponding to the position for the regional area determined is extracted, to obtain the classification of user's area-of-interest.
6. the method according to claim 1, wherein the training process of the detection model the following steps are included:
Initialize the parameter of the neural network;
The sample image is input to the neural network, passes through the position of propagated forward output regional area, corresponding point
Class;
Example segmentation is carried out to the regional area using mask branch;
Position based on the regional area, corresponding classification and example segmentation as a result, mark with the sample image
Value is compared, and optimizes the parameter of the neural network.
7. according to the method described in claim 6, it is characterized in that, the loss function of the neural network is regional area position
Loss function, the loss function of corresponding classification and the sum of the loss function of example segmentation.
8. a kind of searcher of local message characterized by comprising
First obtains module, obtains each part letter of each picture in the picture and picture library to be checked with markup information
Breath, the local message include the position of regional area, corresponding classification, example segmentation result in regional area;Wherein, described
Markup information is used to indicate the location information of user's area-of-interest;
Input module obtains each in the picture to be checked for the picture to be checked to be input to the detection model
The position of regional area, corresponding classification, example segmentation result in each regional area;
Determining module determines the use for the position using each regional area in the markup information and picture to be checked
The classification of family area-of-interest;
Extraction module, for from each picture, extracting correspondence respectively in the picture to be checked and the picture library
In example segmentation result in the regional area of determining classification;
Search module, for based on the example segmentation result out of regional area that taken out in the picture to be checked, described
It is scanned in example segmentation result in the regional area taken out in each picture in picture library, it is described to be checked to obtain
The search result of picture.
9. a kind of electronic equipment characterized by comprising
Memory and processor communicate with each other connection, are stored in the memory between the memory and the processor
Computer instruction, the processor is by executing the computer instruction, thereby executing of any of claims 1-7
The searching method of local message.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer to refer to
It enables, the search that the computer instruction is used to that the computer perform claim to be made to require local message described in any one of 1-7
Method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910120165.5A CN109960742B (en) | 2019-02-18 | 2019-02-18 | Local information searching method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910120165.5A CN109960742B (en) | 2019-02-18 | 2019-02-18 | Local information searching method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109960742A true CN109960742A (en) | 2019-07-02 |
CN109960742B CN109960742B (en) | 2021-11-05 |
Family
ID=67023836
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910120165.5A Active CN109960742B (en) | 2019-02-18 | 2019-02-18 | Local information searching method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109960742B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110490272A (en) * | 2019-09-05 | 2019-11-22 | 腾讯音乐娱乐科技(深圳)有限公司 | Image content similarity analysis method, apparatus and storage medium |
CN110659372A (en) * | 2019-09-24 | 2020-01-07 | 阿里巴巴集团控股有限公司 | Picture input and access method, device and equipment |
CN110689527A (en) * | 2019-09-18 | 2020-01-14 | 北京航空航天大学 | Method, device and equipment for detecting installation state of aircraft cable bracket |
CN110866532A (en) * | 2019-11-07 | 2020-03-06 | 浙江大华技术股份有限公司 | Object matching method and device, storage medium and electronic device |
CN111352814A (en) * | 2020-02-24 | 2020-06-30 | 广州视源电子科技股份有限公司 | Operation abnormal behavior recognition method and device, storage medium and electronic equipment |
CN111531546A (en) * | 2020-05-22 | 2020-08-14 | 济南浪潮高新科技投资发展有限公司 | Robot pose estimation method, device, equipment and storage medium |
CN112084364A (en) * | 2020-09-11 | 2020-12-15 | 苏州科达科技股份有限公司 | Object analysis method, local image search method, device, and storage medium |
CN112308106A (en) * | 2019-11-15 | 2021-02-02 | 北京京邦达贸易有限公司 | Image labeling method and system |
CN112907605A (en) * | 2021-03-19 | 2021-06-04 | 南京大学 | Data enhancement method for instance segmentation |
CN113704525A (en) * | 2020-05-22 | 2021-11-26 | 北京搜狗科技发展有限公司 | Searching method and device and electronic equipment |
TWI770507B (en) * | 2019-08-29 | 2022-07-11 | 大陸商深圳市商湯科技有限公司 | Method and device for retrieving an image and computer readable storage medium |
WO2023272659A1 (en) * | 2021-06-30 | 2023-01-05 | 东莞市小精灵教育软件有限公司 | Method and apparatus for recognizing cover image, storage medium, and recognition device |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020136449A1 (en) * | 2001-01-20 | 2002-09-26 | Samsung Electronics Co., Ltd. | Apparatus and method for extracting object based on feature matching between segmented regions in images |
CN101216841A (en) * | 2008-01-14 | 2008-07-09 | 南京搜拍信息技术有限公司 | Interactive type image search system and method |
CN102129700A (en) * | 2011-03-11 | 2011-07-20 | 上海海事大学 | Infrared simulated image platform under ocean background and image generation method thereof |
CN102968799A (en) * | 2012-12-12 | 2013-03-13 | 北京航空航天大学 | Integral image-based quick ACCA-CFAR SAR (Automatic Censored Cell Averaging-Constant False Alarm Rate Synthetic Aperture Radar) image target detection method |
US20140314313A1 (en) * | 2013-04-17 | 2014-10-23 | Yahoo! Inc. | Visual clothing retrieval |
US9031948B1 (en) * | 2011-07-06 | 2015-05-12 | Shawn B. Smith | Vehicle prediction and association tool based on license plate recognition |
CN105320705A (en) * | 2014-08-05 | 2016-02-10 | 北京大学 | Retrieval method and device for similar vehicle |
CN105373586A (en) * | 2015-10-14 | 2016-03-02 | 浙江宇视科技有限公司 | Vehicle inquiry method and device |
CN106682092A (en) * | 2016-11-29 | 2017-05-17 | 深圳市华尊科技股份有限公司 | Target retrieval method and terminal |
CN107341805A (en) * | 2016-08-19 | 2017-11-10 | 北京市商汤科技开发有限公司 | Background segment and network model training, image processing method and device before image |
CN108846314A (en) * | 2018-05-08 | 2018-11-20 | 天津大学 | A kind of food materials identification system and food materials discrimination method based on deep learning |
CN108921850A (en) * | 2018-04-16 | 2018-11-30 | 博云视觉(北京)科技有限公司 | A kind of extracting method of the image local feature based on image Segmentation Technology |
-
2019
- 2019-02-18 CN CN201910120165.5A patent/CN109960742B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020136449A1 (en) * | 2001-01-20 | 2002-09-26 | Samsung Electronics Co., Ltd. | Apparatus and method for extracting object based on feature matching between segmented regions in images |
CN101216841A (en) * | 2008-01-14 | 2008-07-09 | 南京搜拍信息技术有限公司 | Interactive type image search system and method |
CN102129700A (en) * | 2011-03-11 | 2011-07-20 | 上海海事大学 | Infrared simulated image platform under ocean background and image generation method thereof |
US9031948B1 (en) * | 2011-07-06 | 2015-05-12 | Shawn B. Smith | Vehicle prediction and association tool based on license plate recognition |
CN102968799A (en) * | 2012-12-12 | 2013-03-13 | 北京航空航天大学 | Integral image-based quick ACCA-CFAR SAR (Automatic Censored Cell Averaging-Constant False Alarm Rate Synthetic Aperture Radar) image target detection method |
US20140314313A1 (en) * | 2013-04-17 | 2014-10-23 | Yahoo! Inc. | Visual clothing retrieval |
CN105320705A (en) * | 2014-08-05 | 2016-02-10 | 北京大学 | Retrieval method and device for similar vehicle |
CN105373586A (en) * | 2015-10-14 | 2016-03-02 | 浙江宇视科技有限公司 | Vehicle inquiry method and device |
CN107341805A (en) * | 2016-08-19 | 2017-11-10 | 北京市商汤科技开发有限公司 | Background segment and network model training, image processing method and device before image |
CN106682092A (en) * | 2016-11-29 | 2017-05-17 | 深圳市华尊科技股份有限公司 | Target retrieval method and terminal |
CN108921850A (en) * | 2018-04-16 | 2018-11-30 | 博云视觉(北京)科技有限公司 | A kind of extracting method of the image local feature based on image Segmentation Technology |
CN108846314A (en) * | 2018-05-08 | 2018-11-20 | 天津大学 | A kind of food materials identification system and food materials discrimination method based on deep learning |
Non-Patent Citations (2)
Title |
---|
付伟 等: "基于深度学习的监控视频目标检索", 《无线电工程》 * |
吴金亮 等: "基于Mask R-CNN的舰船目标检测研究", 《无线电工程》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI770507B (en) * | 2019-08-29 | 2022-07-11 | 大陸商深圳市商湯科技有限公司 | Method and device for retrieving an image and computer readable storage medium |
CN110490272A (en) * | 2019-09-05 | 2019-11-22 | 腾讯音乐娱乐科技(深圳)有限公司 | Image content similarity analysis method, apparatus and storage medium |
CN110490272B (en) * | 2019-09-05 | 2022-10-21 | 腾讯音乐娱乐科技(深圳)有限公司 | Image content similarity analysis method and device and storage medium |
CN110689527B (en) * | 2019-09-18 | 2021-08-24 | 北京航空航天大学 | Method, device and equipment for detecting installation state of aircraft cable bracket |
CN110689527A (en) * | 2019-09-18 | 2020-01-14 | 北京航空航天大学 | Method, device and equipment for detecting installation state of aircraft cable bracket |
CN110659372A (en) * | 2019-09-24 | 2020-01-07 | 阿里巴巴集团控股有限公司 | Picture input and access method, device and equipment |
CN110866532A (en) * | 2019-11-07 | 2020-03-06 | 浙江大华技术股份有限公司 | Object matching method and device, storage medium and electronic device |
CN110866532B (en) * | 2019-11-07 | 2022-12-30 | 浙江大华技术股份有限公司 | Object matching method and device, storage medium and electronic device |
CN112308106A (en) * | 2019-11-15 | 2021-02-02 | 北京京邦达贸易有限公司 | Image labeling method and system |
CN111352814A (en) * | 2020-02-24 | 2020-06-30 | 广州视源电子科技股份有限公司 | Operation abnormal behavior recognition method and device, storage medium and electronic equipment |
CN113704525A (en) * | 2020-05-22 | 2021-11-26 | 北京搜狗科技发展有限公司 | Searching method and device and electronic equipment |
CN111531546A (en) * | 2020-05-22 | 2020-08-14 | 济南浪潮高新科技投资发展有限公司 | Robot pose estimation method, device, equipment and storage medium |
CN112084364A (en) * | 2020-09-11 | 2020-12-15 | 苏州科达科技股份有限公司 | Object analysis method, local image search method, device, and storage medium |
CN112907605A (en) * | 2021-03-19 | 2021-06-04 | 南京大学 | Data enhancement method for instance segmentation |
CN112907605B (en) * | 2021-03-19 | 2023-11-17 | 南京大学 | Data enhancement method for instance segmentation |
WO2023272659A1 (en) * | 2021-06-30 | 2023-01-05 | 东莞市小精灵教育软件有限公司 | Method and apparatus for recognizing cover image, storage medium, and recognition device |
Also Published As
Publication number | Publication date |
---|---|
CN109960742B (en) | 2021-11-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109960742A (en) | The searching method and device of local message | |
Liu et al. | ABNet: Adaptive balanced network for multiscale object detection in remote sensing imagery | |
CN108520229A (en) | Image detecting method, device, electronic equipment and computer-readable medium | |
JP2019514123A (en) | Remote determination of the quantity stored in containers in geographical areas | |
CN111027481B (en) | Behavior analysis method and device based on human body key point detection | |
CN111612822B (en) | Object tracking method, device, computer equipment and storage medium | |
CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
CN111160111B (en) | Human body key point detection method based on deep learning | |
CN108960115A (en) | Multi-direction Method for text detection based on angle point | |
CN109299305A (en) | A kind of spatial image searching system based on multi-feature fusion and search method | |
CN110399826B (en) | End-to-end face detection and identification method | |
CN110580446A (en) | Behavior semantic subdivision understanding method, system, computer device and medium | |
US20190332625A1 (en) | Apparatus and method for searching for building based on image and method of constructing building search database for image-based building search | |
CN111311611A (en) | Real-time three-dimensional large-scene multi-object instance segmentation method | |
CN113537180A (en) | Tree obstacle identification method and device, computer equipment and storage medium | |
Sun et al. | Marine ship instance segmentation by deep neural networks using a global and local attention (GALA) mechanism | |
CN105631849B (en) | The change detecting method and device of target polygon | |
Li et al. | Learning to holistically detect bridges from large-size vhr remote sensing imagery | |
CN110472092B (en) | Geographical positioning method and system of street view picture | |
CN113569600A (en) | Method and device for identifying weight of object, electronic equipment and storage medium | |
Santos et al. | RECOGNIZING AND EXPLORING AZULEJOS ON HISTORIC BUILDINGS’FACADES BY COMBINING COMPUTER VISION AND GEOLOCATION IN MOBILE AUGMENTED REALITY APPLICATIONS | |
CN113743487A (en) | Enhanced remote sensing image target detection method and system | |
Bo et al. | Improved YOLOv5 in Remote Sensing Slender and Rotating Target Detection | |
Zhang et al. | An Improved Yolov3 Object Detection Algorithm for UAV Aerial Images | |
Han et al. | Crowded pedestrian detection with optimal bounding box relocation |
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 |