CN106777349A - Face retrieval system and method based on deep learning - Google Patents
Face retrieval system and method based on deep learning Download PDFInfo
- Publication number
- CN106777349A CN106777349A CN201710032351.4A CN201710032351A CN106777349A CN 106777349 A CN106777349 A CN 106777349A CN 201710032351 A CN201710032351 A CN 201710032351A CN 106777349 A CN106777349 A CN 106777349A
- Authority
- CN
- China
- Prior art keywords
- face
- retrieval
- characteristic vector
- image
- facial image
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Library & Information Science (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of face retrieval method and system based on deep learning, comprise the steps of：The training of retrieval model：Using caffe deep learning framework training patterns, the method optimizing model parameter declined using gradient draws retrieval model；The extraction of face characteristic：First proprietary picture in training set is fed in neutral net, feed forward process is carried out, first full articulamentum and second output of full articulamentum is preserved；Face registration：Fully extract characteristic vector and Hash coding of the registrant under different illumination, different attitudes；Face retrieval：Extract the Hash coding and characteristic vector of facial image to be retrieved, then it is compared with the picture in training set, the retrieval mode that coarse search and characteristic vector examining rope are combined is encoded using Hash, by obtaining most like facial image after coarse search and examining rope, and export corresponding Similarity value.
Description
Technical field
The present invention relates to machine vision and face retrieval field, and in particular to face retrieval system based on deep learning and
Method.
Background technology
After krizhevsky et al. proposes based on the theoretical depth convolutional neural networks Alexnet of deep learning, figure
As identification field enters new epoch.The difference of extraction image of the depth convolutional neural networks by convolution algorithm from the superficial to the deep
The feature of level, and learnt the parameter that network automatically adjusts convolution kernel by learning algorithm, in image classification and knowledge
Remarkable result is had been achieved on not.
Main pretreatment, three step groups of feature extraction and characteristic matching by facial image of current face retrieval system
Into.Preprocessing process needs to detect face that this part of technology is more ripe, repeats no more here.Feature is carried
Take the content information of image itself is extracted exactly from image, complete the quantization of image, user is carried out accordingly
Image retrieval.Conventional characteristics of image has SIFT, SURF and PCA-SIFT etc. at present.But different illumination, personage's attitude,
Under expression influence, face information is changed greatly in pattern；Some personages wear ornament, such as beard, glasses due to face
Deng causing face information to distort；The shooting angle of facial image is often varied, and same face is in different angle shot bars
The image difference obtained under part is very big.These manual features can not characterize well it is above-mentioned in the case of image.Additionally, above-mentioned carry
The characteristic dimension got is often higher, it is easy to cause dimension disaster, so can significantly relatively low face retrieval speed, it is impossible to it is real
When return inquiry result.
The method that the method most close with the present invention has Wang Yun [1] et al. propositions, its feature for extracting facial image first,
Feature is mapped to Hamming space from Euclidean space by setting up projection matrix and realizes dimensionality reduction, compiled using improved many bits
Code method is encoded to the feature after dimensionality reduction and is generated image signatures, and signature is weighed with manhatton distance substitution Hamming distance
Between similarity；Then inverted index table is built according to every signature of image in image library；It is high finally by inverted index table
The image close with query image is searched as retrieval result in effect ground.
[1] research of Wang Yun magnanimity facial image method for quickly retrieving and realization [D] China Science ＆ Technology University, 2014.
[2] Wu X, He R, Sun Z, et al.A Light CNN for Deep Face Representation
With Noisy Labels [J] .Computer Science, 2016.
[3] Sun Y, Wang X, Tang X.Deep Learning Face Representation from
Predicting 10,000Classes[J].2014：1891-1898.
[4] Lin K, Yang HF, Hsiao JH, et al.Deep learning of binary hash codes
for fast image retrieval[C]//IEEE Conference on Computer Vision and Pattern
Recognition Workshops.2015：27-35.
[5] Wang Liqing, yellow pine outstanding person are based on image retrieval algorithm [J] software guides of multiple dimensioned convolutional neural networks,
2016,15 (2)：38-40.
[6] Yang Zhiguang, the extra large boat of Chinese mugwort are based on facial image retrieval and relevant feedback [J] the .Acta Automatica of cluster
Sinica, 2008,34 (9)：1033-1039.
[7] Liu Yang are based on image retrieval technologies research [D] the Central China University of Science and Technology of deep learning, 2015.
The content of the invention
Shortcoming and deficiency it is an object of the invention to overcome prior art, for these problems, the system has used volume
The feature that product neutral net is extracted, it is representational strong, there is good sign ability to the face under varying environment；Dimension is low, can
To accelerate retrieval rate.Additionally, the system employs the characteristic matching mode of coarse search and examining rope, face figure is greatly accelerated
The retrieval rate of picture.
The inventive method is related to machine vision and face retrieval, i.e., by camera capture images, then using convolution god
Through network extraction characteristics of image, the process of the image being similar in database is finally found using these features.
A kind of face retrieval method based on deep learning, it is characterised in that comprise the steps of：
The training of retrieval model：Use caffe deep learning framework training patterns, the method optimizing mould declined using gradient
Shape parameter, draws retrieval model；
The extraction of face characteristic：First proprietary picture in training set is fed in neutral net, feed forward process is carried out, protected
Deposit the output of first full articulamentum and second full articulamentum；
Face registration：Fully extract characteristic vector and Hash coding of the registrant under different illumination, different attitudes；
Face retrieval：Extract the Hash coding and characteristic vector of facial image to be retrieved, then by its with training set in
Picture compare, the retrieval mode that coarse search and characteristic vector examining rope are combined is encoded using Hash, by coarse search
With most like facial image is obtained after examining rope, and export corresponding Similarity value.
It is used for examining rope using first output of full articulamentum as the characteristic vector of facial image in the step 2, will
The output of second full articulamentum is encoded for coarse search as the Hash of facial image.
A kind of face retrieval system based on the face retrieval method, can be applied to recognition of face, stranger's report
Alert field.
The beneficial effects of the invention are as follows：Hidden layer by the use of the disaggregated model of deep-neural-network is exported as facial image
Character representation.Additionally, the present invention also extracts the Hash coding of facial image from convolution net, face figure is carried out first with it
The coarse search of picture, obtains some candidate similar face images.Examining rope is carried out with the characteristic vector of face again, obtains final similar
Facial image.It is demonstrated experimentally that the search method that the present invention is used has, and speed is fast, the advantages of retrieval quality is high.This inspection simultaneously
Cable system can be used for the fields such as recognition of face, stranger's alarm.
Brief description of the drawings
Fig. 1 is the flow chart of face search method；
Fig. 2 is face retrieval model figure；
Fig. 3 a are characterized input figure；
Fig. 3 b are first characteristic pattern of convolutional layer；
Fig. 3 c are the 4th characteristic pattern of convolutional layer；
Fig. 3 d are the 5th characteristic pattern of convolutional layer；
Fig. 4 is face register flow path figure；
Fig. 5 a are image to be retrieved；
The Candidate Set that Fig. 5 b are obtained for coarse search；
Fig. 5 c are the similar image that examining rope is obtained；
Fig. 6 is face testing result；
Fig. 7 is retrieval result；
Fig. 8 is the FB(flow block) of face searching system.
Specific embodiment
Below in conjunction with accompanying drawing, the present invention is described in detail：
A kind of face retrieval method of deep learning, Fig. 1 is the flow chart of face search method, mainly including following step
Suddenly：
First, the training of retrieval model
The retrieval model that the system is used is that the lightened CNN models based on Wu Xiang are trained, different
It is to insert Hash coding layer in the layer second from the bottom of network.Final model is had altogether including 14 convolutional layers and 3 full connections
Layer.Convolutional layer therein is used for extracting step by step the feature of facial image, and full articulamentum is used for final classification.Because training nerve
Network needs substantial amounts of data, so combining our gathered data collection as training we used CASIA-WebFace
Data.Final training data has 494414 standardized face's pictures comprising 10228 people altogether.The system has used caffe
Deep learning framework training pattern, the method optimizing model parameter declined using gradient, the bacth of training set is set to 64, checking
The batch of collection is set to 32, and the renewal strategy use " inv " of learning rate is
Lr=base_lr* (1+gamma*iter) ^ (- power)
Wherein lr is the learning rate of current iteration, and base_lr is that basic learning rate is set to 0.01, gamma and power and is all
Constant, is set to 0.000005 and 0.75.Model has iteration altogether 5,000,000 times, and 98.7% is finally reached on test set
Accuracy rate.This illustrates that our model has certain practicality.
2nd, the extraction of face characteristic
If being completed, it is necessary to extract feature representation of the dried layer as facial image in neutral net after the completion of model training
Image vector, is easy to face retrieval below.
Neutral net is to extract face characteristic step by step, i.e., wait until face office from the characteristics of image such as gradient, color block of bottom
Portion's feature arrives the global characteristics of face again.High level is more arrived, the feature of extraction is more abstract, the sign ability of feature is stronger.Each layer is carried
The characteristic pattern got is as shown in Figure 3.In order to complete face retrieval system, it is necessary to proprietary picture in training set first is fed into god
In through network, feed forward process is carried out, preserve first full articulamentum and second output of full articulamentum.In this searching system
In, it is used for examining rope using first output of full articulamentum as the characteristic vector of facial image, by second full articulamentum
Output is encoded for coarse search as the Hash of facial image.
3rd, face registration
The system is mainly used in above safety protection robot.Safety protection robot is gone on patrol in company, it is necessary to be judged certain
People is the employee of our company.Need to extract its corresponding characteristic vector for the employee that newly arrives and Hash is encoded.Concrete operations
Process is as shown in Figure 4.
In order to improve the success rate of retrieval, it is necessary to fully extract feature of the registrant under different illumination, different attitudes to
Amount and Hash coding.
4th, face retrieval
After the Hash coding and characteristic vector of face images in training set have been preserved, this searching system is using Kazakhstan
The retrieval mode that the coarse search of uncommon coding and characteristic vector examining rope are combined.
In nearest neighbor search problem, a facial image to be checked is given, model needs to return to similar time
Select image.Here it is near that similar refers to seeing in appearance.Traditional method is exactly that every image (ratio is represented with characteristic vector
Such as the output of a certain layer in neutral net), then by image in calculating inquiry image to be checked and database in feature space
In Euclidean distance, and according to distance order from small to large, the image similar with image to be checked in returned data storehouse.It is above-mentioned
Although method looks very easy and effective, but as the image in face database is more and more, the short slab of this method also embodies
Incisively and vividly --- memory space consumption is big, and retrieval rate is slow.Specifically, it is assumed that characterized using the characteristic vector of 4096 dimensions
One facial image, then represent that 1,000,000 images are accomplished by the memory space (single precision floating datum) of about 156B, and count
Every distance of image in query image and database is calculated, is then needed 8192 sub-additions to operate and 4096 multiplication is operated, traversal
Complete all of 1,000,000 images are returned again to if result, overlong time, it is difficult to put into actually used.
In hash algorithm, sample is expressed as a string of binary-codings of regular length, usually using 0,1 or -1,1 table
Show each of which position so that similar sample has similar two-value code, thus can measure two-value using Hamming distance
Similitude between code.The retrieval rate of image can be so greatly speeded up, but hash algorithm equally has the shortcomings that oneself such as
Characterize ability etc..
This searching system makes full use of Hash to encode the advantage that calculating speed is fast and characteristic vector sign ability is strong, and utilizes
The Hash coding of the method study image of deep learning, it is to avoid hand-designed coding it is cumbersome.During actual use,
For image to be checked, by model extraction, its Hash is encoded and characteristic vector first, is then compiled using the Hash of image
Code calculates the Hamming distance of image to be checked and image in face database, and 50 nearest images of selected distance are used as Candidate Set.Most
The cosine similarity (as shown in the figure) of image to be checked and Candidate Set image is calculated using the characteristic vector of image afterwards, is chosen similar
10 maximum images are spent as final output image.Detailed process is as shown in Figure 5.
The dimension of n representative features vector, AiRepresent the characteristic vector of image to be checked, BiRepresent the spy of image in Candidate Set
Levy vector.
It is assumed that needing to return to 10 similar facial images, compared to directly time-consuming 0.27414s is retrieved using characteristic vector, adopt
Time-consuming 0.005175s (coarse search 0.003175s, examining rope 0.002s) is only needed with this retrieval mode.As can be seen here, this retrieval side
Formula substantially increases retrieval rate.
It is below the process of actual verification：
Operation program, a two field picture is captured and by Face datection algorithm by camera, detects facial image, is such as schemed
Shown in 6.
After by facial image gray processing, it is fed in the model for training, extracts the Hash coding and feature of facial image
Vector.Then it is compared with the picture in training set, by obtaining 10 most like people after coarse search and examining rope
Face image, and export corresponding Similarity value.As shown in Figure 7.
Fig. 8 is the FB(flow block) of face searching system.Mainly include Face datection, image preprocessing, face characteristic is extracted,
Hash encodes five steps of coarse search and characteristic vector examining rope.
The beneficial effects of the invention are as follows：
For the current slow-footed situation of face retrieval system retrieval, it is combined using Hash coding and characteristic vector matching
Mode carry out face retrieval, substantially increase the retrieval rate of image.
For the current low-quality situation of face retrieval system retrieval, using deep learning method, using convolutional Neural net
The characteristic vector of network model learning image, dimension is low, representational strong.
For the cumbersome of the manual demarcation of current Hash coding, using deep learning method, using convolutional neural networks model
The Hash coding of automatic study image.
For the diversity of scene residing for face, in different illumination, different attitudes gather facial image under different illumination
Hash is encoded and characteristic vector.
Above-described embodiment is the present invention preferably implementation method, but embodiments of the present invention are not by above-described embodiment
Limitation, it is other it is any without departing from Spirit Essence of the invention and the change, modification, replacement made under principle, combine, simplification,
Equivalent substitute mode is should be, is included within protection scope of the present invention.
Claims (3)
1. a kind of face retrieval method based on deep learning, it is characterised in that comprise the steps of：
First, the training of retrieval model：Use caffe deep learning framework training patterns, the method optimizing mould declined using gradient
Shape parameter, draws retrieval model；
2nd, the extraction of face characteristic：First proprietary picture in training set is fed in neutral net, feed forward process is carried out, protected
Deposit the output of first full articulamentum and second full articulamentum；
3rd, face registration：Fully extract characteristic vector and Hash coding of the registrant under different illumination, different attitudes；
4th, face retrieval：Extract the Hash coding and characteristic vector of facial image to be retrieved, then by its with training set in
Picture is compared, and the retrieval mode that coarse search and characteristic vector examining rope are combined is encoded using Hash, by coarse search and
Most like facial image is obtained after examining rope, and exports corresponding Similarity value.
2. a kind of face retrieval method according to claim 1, it is characterised in that connect first entirely in the step 2
The output for connecing layer is used for examining rope as the characteristic vector of facial image, using second output of full articulamentum as facial image
Hash encode for coarse search.
3. a kind of face retrieval system for applying face retrieval method as claimed in claim 1, it is characterised in that applied
In recognition of face, stranger's alarm field.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710032351.4A CN106777349A (en) | 2017-01-16 | 2017-01-16 | Face retrieval system and method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710032351.4A CN106777349A (en) | 2017-01-16 | 2017-01-16 | Face retrieval system and method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106777349A true CN106777349A (en) | 2017-05-31 |
Family
ID=58946066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710032351.4A Pending CN106777349A (en) | 2017-01-16 | 2017-01-16 | Face retrieval system and method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106777349A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633258A (en) * | 2017-08-21 | 2018-01-26 | 北京精密机电控制设备研究所 | A kind of deep learning identifying system and method based on feed-forward character extraction |
| CN107742116A (en) * | 2017-11-13 | 2018-02-27 | 湖南超能机器人技术有限公司 | A kind of infant emotion change detection and knowledge method for distinguishing |
| CN107958247A (en) * | 2018-01-17 | 2018-04-24 | 百度在线网络技术（北京）有限公司 | Method and apparatus for facial image identification |
| CN107992611A (en) * | 2017-12-15 | 2018-05-04 | 清华大学 | The high dimensional data search method and system of hash method are distributed based on Cauchy |
| CN108345654A (en) * | 2018-01-23 | 2018-07-31 | 南京邮电大学 | A kind of image Hash search method based on semi-supervised ladder network |
| CN108363771A (en) * | 2018-02-08 | 2018-08-03 | 杭州电子科技大学 | A kind of image search method towards public security investigation application |
| CN109359530A (en) * | 2018-09-12 | 2019-02-19 | 中国人民解放军国防科技大学 | Intelligent video monitoring method and device |
| CN110059634A (en) * | 2019-04-19 | 2019-07-26 | 山东博昂信息科技有限公司 | A kind of large scene face snap method |
| CN110175248A (en) * | 2019-04-04 | 2019-08-27 | 中国科学院信息工程研究所 | A kind of Research on face image retrieval and device encoded based on deep learning and Hash |
| CN110378385A (en) * | 2019-06-20 | 2019-10-25 | 安徽省农业科学院畜牧兽医研究所 | A kind of beef texture automatic measure grading method, system, device and storage medium |
| CN111143597A (en) * | 2019-12-13 | 2020-05-12 | 浙江大华技术股份有限公司 | Image retrieval method, terminal and storage device |
| WO2020098296A1 (en) * | 2018-11-15 | 2020-05-22 | ***股份有限公司 | Image retrieval method and device |
| CN111401291A (en) * | 2020-03-24 | 2020-07-10 | 三一重工股份有限公司 | Stranger identification method and device |
| CN111598012A (en) * | 2020-05-19 | 2020-08-28 | 恒睿(重庆)人工智能技术研究院有限公司 | Picture clustering management method, system, device and medium |
| CN114329029A (en) * | 2021-10-28 | 2022-04-12 | 腾讯科技（深圳）有限公司 | Object retrieval method, device, equipment and computer storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104834748A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Image retrieval method utilizing deep semantic to rank hash codes |
| CN104899579A (en) * | 2015-06-29 | 2015-09-09 | 小米科技有限责任公司 | Face recognition method and face recognition device |
| CN105279554A (en) * | 2015-09-29 | 2016-01-27 | 东方网力科技股份有限公司 | Depth neural network training method and device based on Hash coding layer |
| CN105718889A (en) * | 2016-01-21 | 2016-06-29 | 江南大学 | Human face identity recognition method based on GB(2D)2PCANet depth convolution model |
-
2017
- 2017-01-16 CN CN201710032351.4A patent/CN106777349A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104834748A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Image retrieval method utilizing deep semantic to rank hash codes |
| CN104899579A (en) * | 2015-06-29 | 2015-09-09 | 小米科技有限责任公司 | Face recognition method and face recognition device |
| CN105279554A (en) * | 2015-09-29 | 2016-01-27 | 东方网力科技股份有限公司 | Depth neural network training method and device based on Hash coding layer |
| CN105718889A (en) * | 2016-01-21 | 2016-06-29 | 江南大学 | Human face identity recognition method based on GB(2D)2PCANet depth convolution model |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633258B (en) * | 2017-08-21 | 2020-04-10 | 北京精密机电控制设备研究所 | Deep learning identification system and method based on feedforward feature extraction |
| CN107633258A (en) * | 2017-08-21 | 2018-01-26 | 北京精密机电控制设备研究所 | A kind of deep learning identifying system and method based on feed-forward character extraction |
| CN107742116A (en) * | 2017-11-13 | 2018-02-27 | 湖南超能机器人技术有限公司 | A kind of infant emotion change detection and knowledge method for distinguishing |
| CN107992611A (en) * | 2017-12-15 | 2018-05-04 | 清华大学 | The high dimensional data search method and system of hash method are distributed based on Cauchy |
| CN107958247A (en) * | 2018-01-17 | 2018-04-24 | 百度在线网络技术（北京）有限公司 | Method and apparatus for facial image identification |
| CN108345654A (en) * | 2018-01-23 | 2018-07-31 | 南京邮电大学 | A kind of image Hash search method based on semi-supervised ladder network |
| CN108363771A (en) * | 2018-02-08 | 2018-08-03 | 杭州电子科技大学 | A kind of image search method towards public security investigation application |
| CN108363771B (en) * | 2018-02-08 | 2020-05-01 | 杭州电子科技大学 | Image retrieval method for public security investigation application |
| CN109359530A (en) * | 2018-09-12 | 2019-02-19 | 中国人民解放军国防科技大学 | Intelligent video monitoring method and device |
| CN109359530B (en) * | 2018-09-12 | 2022-01-25 | 中国人民解放军国防科技大学 | Intelligent video monitoring method and device |
| WO2020098296A1 (en) * | 2018-11-15 | 2020-05-22 | ***股份有限公司 | Image retrieval method and device |
| CN110175248B (en) * | 2019-04-04 | 2021-08-31 | 中国科学院信息工程研究所 | Face image retrieval method and device based on deep learning and Hash coding |
| CN110175248A (en) * | 2019-04-04 | 2019-08-27 | 中国科学院信息工程研究所 | A kind of Research on face image retrieval and device encoded based on deep learning and Hash |
| CN110059634A (en) * | 2019-04-19 | 2019-07-26 | 山东博昂信息科技有限公司 | A kind of large scene face snap method |
| CN110378385A (en) * | 2019-06-20 | 2019-10-25 | 安徽省农业科学院畜牧兽医研究所 | A kind of beef texture automatic measure grading method, system, device and storage medium |
| CN111143597A (en) * | 2019-12-13 | 2020-05-12 | 浙江大华技术股份有限公司 | Image retrieval method, terminal and storage device |
| CN111143597B (en) * | 2019-12-13 | 2023-06-20 | 浙江大华技术股份有限公司 | Image retrieval method, terminal and storage device |
| CN111401291A (en) * | 2020-03-24 | 2020-07-10 | 三一重工股份有限公司 | Stranger identification method and device |
| CN111598012A (en) * | 2020-05-19 | 2020-08-28 | 恒睿(重庆)人工智能技术研究院有限公司 | Picture clustering management method, system, device and medium |
| CN111598012B (en) * | 2020-05-19 | 2021-11-12 | 恒睿(重庆)人工智能技术研究院有限公司 | Picture clustering management method, system, device and medium |
| CN114329029A (en) * | 2021-10-28 | 2022-04-12 | 腾讯科技（深圳）有限公司 | Object retrieval method, device, equipment and computer storage medium |
| CN114329029B (en) * | 2021-10-28 | 2024-05-14 | 腾讯科技（深圳）有限公司 | Object retrieval method, device, equipment and computer storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106777349A (en) | Face retrieval system and method based on deep learning | |
| CN109948425B (en) | Pedestrian searching method and device for structure-aware self-attention and online instance aggregation matching | |
| CN107229757B (en) | Video retrieval method based on deep learning and Hash coding | |
| CN111753189B (en) | Few-sample cross-modal hash retrieval common characterization learning method | |
| CN107122375B (en) | Image subject identification method based on image features | |
| CN104090972B (en) | The image characteristics extraction retrieved for D Urban model and method for measuring similarity | |
| CN111709311B (en) | Pedestrian re-identification method based on multi-scale convolution feature fusion | |
| CN108595636A (en) | The image search method of cartographical sketching based on depth cross-module state correlation study | |
| CN108875813B (en) | Three-dimensional grid model retrieval method based on geometric image | |
| CN106126581A (en) | Cartographical sketching image search method based on degree of depth study | |
| CN108304573A (en) | Target retrieval method based on convolutional neural networks and supervision core Hash | |
| CN102254015A (en) | Image retrieval method based on visual phrases | |
| CN110929080A (en) | Optical remote sensing image retrieval method based on attention and generation countermeasure network | |
| CN113343901A (en) | Human behavior identification method based on multi-scale attention-driven graph convolutional network | |
| CN106355210B (en) | Insulator Infrared Image feature representation method based on depth neuron response modes | |
| CN114694185B (en) | Cross-modal target re-identification method, device, equipment and medium | |
| CN108446605B (en) | Double interbehavior recognition methods under complex background | |
| CN108763295A (en) | A kind of video approximate copy searching algorithm based on deep learning | |
| CN112434553A (en) | Video identification method and system based on deep dictionary learning | |
| CN113051984A (en) | Video copy detection method and apparatus, storage medium, and electronic apparatus | |
| CN117011883A (en) | Pedestrian re-recognition method based on pyramid convolution and transducer double branches | |
| CN110442749B (en) | Video frame processing method and device | |
| CN116977663A (en) | Image data processing method, device, equipment and medium | |
| CN106055636B (en) | portable intelligent rock identification method | |
| CN112256895B (en) | Fabric image retrieval method based on multitask learning |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170531 |
|
| WD01 | Invention patent application deemed withdrawn after publication |