CN105654054B - The intelligent video analysis method of study and more visual dictionary models is propagated based on semi-supervised neighbour - Google Patents
The intelligent video analysis method of study and more visual dictionary models is propagated based on semi-supervised neighbour Download PDFInfo
- Publication number
- CN105654054B CN105654054B CN201511022492.5A CN201511022492A CN105654054B CN 105654054 B CN105654054 B CN 105654054B CN 201511022492 A CN201511022492 A CN 201511022492A CN 105654054 B CN105654054 B CN 105654054B
- Authority
- CN
- China
- Prior art keywords
- video
- key
- frame
- feature vector
- visual
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of intelligent video analysis methods that study and more visual dictionary models are propagated based on semi-supervised neighbour, comprising: keeps the key video sequence frame of strategy to extract based on sampling;Key frame is converted into gray level image, averagely segmentation gray level image is several image blocks, generates OM feature vector with the average brightness sequence of each image block;Based on AP clustering video sample, the video submanifold of mutual exclusion is formed;For the video submanifold of AP clustering, determines the corresponding class label of each submanifold, construct more visual dictionaries;According to minimum range decision rule, the class label of video to be checked is judged online;The class label that the more visual dictionaries to shake down rejudge video to be checked is reconstructed using the self-adapting reconstruction study of closed loop feedback.The present invention solves the problems, such as that key-frame extraction cannot take into account efficiency and precision, and enhancing video features pairing approximation overall situation linear change has very strong strong robustness, realizes the accurate intelligent measurement that can be adaptive to video mode variation.
Description
Technical field
The present invention relates to computer graphic image processing technology fields, in particular to a kind of to be based on semi-supervised neighbour's dissemination
Practise the intelligent video analysis method with more visual dictionary models.
Background technique
Universal with the terminal devices such as smart phone, digital camera and internet, the production of video stores and transmits more
It is convenient to add, and more and more people watch video by network, and the life information of oneself is fabricated to video and is uploaded to network point
It enjoys, video traffic has been no longer limited to the services for life industries such as traditional radio and television, amusement film, and is widely used in political affairs
It controls, military affairs, science and education, medical treatment, traffic, the various fields such as safety.In face of " explosive " growth of number of videos, how quick and precisely
Ground retrieval, matching, classification video information face huge challenge, while being also the hot issue of academia's research.
Existing intellectual analysis key technology mainly includes three kinds: key-frame extraction, feature detection and machine learning, key frame
Extract: key frame is one group of image of reflecting video segment main contents, and the extraction method of key frame based on shot segmentation is accurate
Property it is high, but calculation amount is too big, time-consuming serious, and the key-frame extraction based on fixed rate has very high timeliness, but extracts
Keyframe sequence possibly can not accurate description video clip main contents, lack representative;Feature detection: one kind utilizes letter
Clean global characteristics quickly detect video, such as color histogram etc., this kind of technology detection efficiency with higher, but are only capable of
Detect some videos without adding global linear change;The another kind of robustness using higher-dimension local feature enhancing algorithm, example
Such as Scale invariant features transform, fast robust feature, this kind of technology still have the video of some nonlinear changes good
Good detection effect even includes the variation of some backgrounds if visual angle change, Geometrical change, picture are cut, but every frame video
In can all extract hundreds and thousands of a local features, cause the calculation amount of characteristic matching huge, detection efficiency not can guarantee;Engineering
It practises: the method that video analysis generally uses machine learning, but the accuracy of machine learning algorithm remains to be further improved, engineering
Practising sorting algorithm is often influenced by following factor: the training sample of most of machine learning algorithm pairing approximation elliposoidal distribution
This collection has preferable classifying quality, but cannot get good classification performance to the complex distributions structure of spill;In practical application
In inevitably can be comprising the noises such as some isolated points, unknown data or wrong data, these can all influence the quality of sorting algorithm;
It cannot be adaptive to the variation of video environment, identify emerging video type in real time.
Summary of the invention
Aiming at the shortcomings in the prior art, the present invention is provided a kind of propagated based on semi-supervised neighbour and learnt and more visual dictionaries
The intelligent video analysis method of model solves the problems, such as that key video sequence frame cannot take into account efficiency and precision in the prior art, enhancing
The robustness of video features pairing approximation overall situation linear change realizes the accurate intelligence inspection that can be adaptive to video mode variation
It surveys.
According to design scheme provided by the present invention, one kind propagating study and more visual dictionary models based on semi-supervised neighbour
Intelligent video analysis method, comprise the following steps:
Step 1. is directed to video sample, keeps strategy to extract key video sequence frame using sampling;
Step 2. is directed to key video sequence frame, calculates the OM feature vector based on sequential metrics;
Step 3. propagates study to all OM feature vectors progress intelligent clustering using based on semi-supervised neighbour, is formed each
Video submanifold;
Step 4. determines the corresponding class label of each video submanifold, constructs more visual dictionaries, and class label includes unknown
Type video label;
Video to be checked is successively executed the sampling of the use in step 1 and strategy is kept to extract key video sequence frame and step by step 5.
OM feature vector of the calculating based on sequential metrics in 2, and according to more visual dictionaries in step 4, according to minimum range rule
Judge the class label of video to be checked;
If the video number that UNKNOWN TYPE video tab occurs in step 6. is greater than given threshold, using closed loop feedback from
Reconstruct learning method is adapted to, return step 3 reconstructs the more visual dictionaries that can adapt to new environment, further judges wait inspect
Otherwise the class label of frequency terminates.
Above-mentioned, step 1 specifically includes following content: for the video frame of any arrival, extracting the abstract letter of video frame
Breath;Summary info is matched with key feature library, if successful match, which is determined as key video sequence frame, it is no
Then, random sampling is carried out according to Probability p and be determined as key video sequence frame if drawing, otherwise, abandon the video frame.
Above-mentioned, step 2 calculates the OM feature vector based on sequential metrics and specifically includes the following steps:
Key video sequence frame is converted into gray level image by step 2.1;
Gray level image is averagely divided into N number of image block by step 2.2, wherein N=Nx*Ny, wherein Nx represents X-axis side
To image block;Ny represents the image block of Y direction;
Step 2.3, the average brightness value I for calculating each image blockk, i.e.,Wherein, f (x,
It y) is brightness value of the coordinate for the pixel of (x, y), k ∈ [1, N];M, n are the line number and columns of image block;
Step 2.4 is ranked up each image block average brightness value, generates OM feature vector I=[I1,I2,……,IN]。
Above-mentioned, step 3 specifically includes following content:
Step 3.1 is directed to marked video sample SPACE Vl, unmarked video sample SPACE Vnl, extract all videos and close
The OM feature vector of key frame;
Step 3.2 successively judges any two OM feature vector Ii、IjWhether V is belonged tolIf belonging to Vl, and it is marked
Video belongs to same type, then by two OM feature vector Ii、IjThe distance between DijIt is set as maximum value 0;If belonging to Vl, and
Marked video is not belonging to uniform type, then by two Ii、IjThe distance between DijIt is set as minimum value-∞;If Ii、IjIn at least
There is one to belong to Vnl, then I is calculatedi、IjEuclidean distance, distance D as between the twoij;
The distance of n sample point any two OM feature vector in video sample space is stored in matrix by step 3.3
In E;
Step 3.4 carries out clustering based on neighbour's propagation principle, forms the K C={ C that cluster1,…,CK}。
Above-mentioned, step 4 specifically includes following content:
Step 4.1 is directed to any video submanifold Ci, calculates the marked video frame for belonging to any class label lj in Ci
Number Ni;
Step 4.2, using majority vote rule, by the class label l comprising most samples*It is assigned to submanifold Ci;
Step 4.3, the mass center wi, as class label l for calculating video submanifold Ci*Vision code book Wi;
Step 4.4, all vision code books constitute more visual dictionaries.
Above-mentioned, step 5 specifically includes following content:
Video to be checked is kept strategy to extract key video sequence frame by step 5.1 using sampling;
Step 5.2 calculates the OM feature vector O based on sequential metrics;
Step 5.3 calculates OM feature vector O at a distance from vision code book Wi any in more visual dictionaries;
Step 5.4 finds out the corresponding vision code book Wj of minimum range;
Video to be checked is judged to the affiliated submanifold of vision code book Wj by step 5.5.
Above-mentioned, the self-adapting reconstruction learning method of closed loop feedback in step 6 reconstructs the more views that can adapt to new environment
Feel dictionary, specifically include following content:
Step 6.1, the maximum differentiation distance Dmax and threshold value δ of initialization;
Step 6.2 is directed to key frame of video OM feature vector Oi, and it is whether big at a distance from all vision code books to compare Oi
In Dmax;
If step 6.3, Oi are all larger than Dmax at a distance from all vision code books, which is judged to UNKNOWN TYPE video;
Otherwise, jump procedure 5.3 executes;
If step 6.4, UNKNOWN TYPE video number are greater than threshold value δ, 3 are gone to step, construct the more views to shake down
Feel dictionary, further judge the type of video to be checked, otherwise, terminates.
Beneficial effects of the present invention:
1, the present invention keeps strategy to extract key frame of video, calculating OM feature vector, based on semi-supervised using sampling is based on
The method that the clustering of AP, more visual dictionaries building based on label mapping and closed loop feedback reconstruct learn, is effectively reduced
The computation complexity of the characteristic matching of massive video, and the precision of visual classification is greatly improved, realize the intelligence point of massive video
Analysis.
2, the present invention is based on the key video sequence frame extraction that sampling keeps strategy, sample process is maintained to facilitate, at guarantee
Reason efficiency is unaffected, and on the other hand the thick matching to key video sequence frame based on known key feature, avoids because of key video sequence frame
Difference caused by extracting;The OM feature vector of sequential metrics does not emphasize the average brightness of each image block, only focuses on each image block
The sequence of average brightness to addition noise, adjustment parameter, the approximate global linear change such as recompiles and has good robust
Property, feature is succinct, is easy to extract;Semi-supervised learning mechanism is introduced, using most of unlabeled exemplars and on a small quantity has exemplar,
The computation complexity for substantially reducing marking video sample knows video class label using a small amount of marker samples automatically, improves and divides
Class precision and accuracy propagate the Clustering Effect that study improves extensive high dimensional feature using neighbour, link vision code book it is same
Justice and ambiguity problem;If occurring unknown video type in video environment, intelligence can be known by way of closed loop feedback reconstruct
Video mode that Chu be not new can adapt to the accurate intelligent measurement changed in video mode.
Detailed description of the invention:
Fig. 1 is flow diagram of the invention;
Fig. 2 is the key video sequence frame extraction flow diagram of the invention that strategy is kept based on sampling;
Fig. 3 is the flow diagram of calculating OM feature vector of the invention;
Fig. 4 is the intelligent clustering flow diagram of the invention based on semi-supervised AP;
Fig. 5 is the more visual dictionary flow diagrams of building of the invention;
Fig. 6 is the flow diagram for the class label that foundation minimum range rule of the invention judges video to be checked;
Fig. 7 is the self-adapting reconstruction learning method flow diagram of the invention based on closed loop feedback.
Specific embodiment:
The present invention is described in further detail with technical solution with reference to the accompanying drawing, and detailed by preferred embodiment
Describe bright embodiments of the present invention in detail, but embodiments of the present invention are not limited to this.
Embodiment one, a kind of shown in Figure 1, intelligence for propagating study and more visual dictionary models based on semi-supervised neighbour
Video analysis method, comprises the following steps:
Step 1. is directed to video sample, keeps strategy to extract key video sequence frame using sampling;
Step 2. is directed to key video sequence frame, calculates the OM feature vector based on sequential metrics;
Step 3. propagates study to all OM feature vectors progress intelligent clustering using based on semi-supervised neighbour, is formed each
Video submanifold;
Step 4. determines the corresponding class label of each video submanifold, constructs more visual dictionaries, and class label includes unknown
Type video label;
Video to be checked is successively executed the sampling of the use in step 1 and strategy is kept to extract key video sequence frame and step by step 5.
OM feature vector of the calculating based on sequential metrics in 2, and according to more visual dictionaries in step 4, according to minimum range rule
Judge the class label of video to be checked;
Step 6. when video number that UNKNOWN TYPE video tab occur is greater than given threshold, using closed loop feedback from
Reconstruct learning method is adapted to, return step 3 reconstructs the more visual dictionaries that can adapt to new environment, otherwise, terminates.
Embodiment two, referring to shown in Fig. 1~7, a kind of propagated based on semi-supervised neighbour is learnt and more visual dictionary models
Intelligent video analysis method, comprises the following steps:
Step 1. is directed to video sample, keeps strategy to extract key video sequence frame using sampling, for the video arbitrarily reached
Frame extracts the summary info of video frame;Summary info is matched with key feature library, if successful match, by the video
Frame is determined as key video sequence frame, otherwise, carries out random sampling according to Probability p and is determined as key video sequence frame if drawing, otherwise, lose
Abandon the video frame, by key video sequence frame extraction be divided into feature slightly match and the video frame fixed cycle sample, if arrival video frame
It is fitted on known key feature, then it is assumed that it is key video sequence frame, is otherwise sampled with fixation probability p;
Step 2. is directed to key video sequence frame, calculates the OM feature vector based on sequential metrics, specifically includes the following steps:
Key video sequence frame is converted into gray level image by step 2.1;
Gray level image is averagely divided into N number of image block by step 2.2, wherein N=Nx*Ny, wherein Nx represents X-axis side
To image block;Ny represents the image block of Y direction;
Step 2.3, the average brightness value I for calculating each image blockk, i.e.,Wherein, f (x,
It y) is brightness value of the coordinate for the pixel of (x, y), k ∈ [1, N];M, n are the line number and columns of image block;
Step 2.4 is ranked up each image block average brightness value, generates OM feature vector I=[I1,I2,……,IN];
Step 3. propagates study to all OM feature vectors progress intelligent clustering using based on semi-supervised neighbour, is formed each
Video submanifold specifically includes following content:
Step 3.1 is directed to marked video sample SPACE Vl, unmarked video sample SPACE Vnl, extract all videos and close
The OM feature vector of key frame;
Step 3.2 successively judges any two OM feature vector Ii、IjWhether V is belonged tolIf belonging to Vl, and it is marked
Video belongs to same type, then by two OM feature vector Ii、IjDistance DijIt is set as maximum value 0;If belonging to Vl, and marked
Note video is not belonging to uniform type, then by two Ii、IjDistance DijIt is set as minimum value-∞;If Ii、IjIn at least one category
In Vnl, then I is calculatedi、IjEuclidean distance Dij;
The distance of any the two of n sample point in video sample space is stored in matrix E by step 3.3;
Step 3.4 carries out clustering based on neighbour's propagation principle, forms the K C={ C that cluster1,…,CK};Step 4. is true
Determine the corresponding class label of each video submanifold, construct more visual dictionaries, class label includes UNKNOWN TYPE video tab, specifically
Include following content:
Step 4.1 is directed to any video submanifold Ci, calculates the marked video frame for belonging to any class label lj in Ci
Number Ni;
Step 4.2, using majority vote rule, by the class label l comprising most samples*It is assigned to submanifold Ci;
Step 4.3, the mass center wi, as class label l for calculating video submanifold Ci*Vision code book Wi;
Step 4.4, all vision code books constitute more visual dictionaries;
Video to be checked is successively executed the sampling of the use in step 1 and strategy is kept to extract key video sequence frame and step by step 5.
OM feature vector of the calculating based on sequential metrics in 2, and according to more visual dictionaries in step 4, according to minimum range rule
Judge the class label of video to be checked, specifically include following content:
Video to be checked is kept strategy to extract key video sequence frame by step 5.1 using sampling;
Step 5.2 calculates the OM feature vector O based on sequential metrics;
Step 5.3 calculates OM feature vector O at a distance from vision code book Wi any in more visual dictionaries;
Step 5.4 finds out the corresponding vision code book Wj of minimum range;
Video to be checked is judged to the affiliated submanifold of vision code book Wj by step 5.5;
Step 6. when video number that UNKNOWN TYPE video tab occur is greater than given threshold, using closed loop feedback from
Reconstruct learning method is adapted to, the more visual dictionaries that can adapt to new environment are reconstructed, specifically includes following content:
Step 6.1, the maximum differentiation distance Dmax and threshold value δ of initialization;
Step 6.2 is directed to key frame of video OM feature vector Oi, and it is whether big at a distance from all vision code books to compare Oi
In Dmax;
If step 6.3, Oi are all larger than Dmax at a distance from all vision code books, which is judged to UNKNOWN TYPE video;
Otherwise, jump procedure 5.3 executes;
If step 6.4, UNKNOWN TYPE video number are greater than threshold value δ, 3 are gone to step, construct the more views to shake down
Feel dictionary, further judge the type of video to be checked, otherwise, terminates.
Difference caused by avoiding key video sequence frame from extracting improves video detection efficiency;To addition noise, adjustment parameter, again
Newly encoded equal approximate global linear change has good robustness;The precision of video analysis model is improved, is improved extensive high
The Clustering Effect of dimensional feature alleviates the synonymy and ambiguity problem of vision code book.
The invention is not limited to above-mentioned specific embodiment, those skilled in the art can also make a variety of variations accordingly,
But it is any all to cover within the scope of the claims with equivalent or similar variation of the invention.
Claims (6)
1. a kind of intelligent video analysis method for being propagated study and more visual dictionary models based on semi-supervised neighbour, feature are existed
In: it comprises the following steps:
Step 1. is directed to video sample, keeps strategy to extract key video sequence frame using sampling;For the video frame arbitrarily reached, mention
Take the summary info of video frame;Summary info is matched with key feature library, if successful match, which is determined
Otherwise random sampling is carried out according to Probability p and is determined as key video sequence frame if drawing, otherwise, abandons the view for key video sequence frame
Frequency frame;
Step 2. is directed to key video sequence frame, calculates the OM feature vector based on sequential metrics;
Step 3. propagates study to all OM feature vectors progress intelligent clustering using based on semi-supervised neighbour, forms each video
Submanifold;
Step 4. determines the corresponding class label of each video submanifold, constructs more visual dictionaries, and class label includes UNKNOWN TYPE
Video tab;
Step 5. successively executes video to be checked keeps strategy to extract key video sequence frame and calculate based on sequential metrics using sampling
OM feature vector the classification mark of video to be checked is judged according to minimum range rule and according to more visual dictionaries in step 4
Label;
If the video number that UNKNOWN TYPE video tab occurs in step 6. is greater than given threshold, the adaptive of closed loop feedback is used
Learning method is reconstructed, return step 3 reconstructs the more visual dictionaries that can adapt to new environment, further judges video to be checked
Otherwise class label terminates.
2. the intelligent video analysis according to claim 1 for propagating study and more visual dictionary models based on semi-supervised neighbour
Method, it is characterised in that: step 2 calculates the OM feature vector based on sequential metrics and specifically includes the following steps:
Key video sequence frame is converted into gray level image by step 2.1;
Gray level image is averagely divided into N number of image block by step 2.2, wherein N=Nx*Ny, wherein Nx represents X-direction
Image block;Ny represents the image block of Y direction;
Step 2.3, the average brightness value I for calculating each image blockk, i.e.,Wherein, f (x, y) is
Coordinate is the brightness value of the pixel of (x, y), k ∈ [1, N];M, n are the line number and columns of image block;Step 2.4, to each image block
Average brightness value is ranked up, and generates OM feature vector I=[I1,I2,……,IN]。
3. the intelligent video analysis according to claim 2 for propagating study and more visual dictionary models based on semi-supervised neighbour
Method, it is characterised in that: step 3 specifically includes following content:
Step 3.1 is directed to marked video sample SPACE Vl, unmarked video sample SPACE Vnl, extract all key frame of video
OM feature vector;
Step 3.2 successively judges any two OM feature vector Ii、IjWhether V is belonged tolIf belonging to Vl, and marked video category
In same type, then by two OM feature vector Ii、IjDistance DijIt is set as maximum value 0;If belonging to Vl, and marked video
It is not belonging to uniform type, then by two Ii、IjDistance DijIt is set as minimum value-∞;If Ii、IjIn at least one belong to Vnl, then
Calculate Ii、IjEuclidean distance Dij;
The distance of any the two of n sample point in video sample space is stored in matrix E by step 3.3;Step 3.4, base
Clustering is carried out in neighbour's propagation principle, forms the K C={ C that cluster1,…,CK}。
4. the intelligent video analysis according to claim 3 for propagating study and more visual dictionary models based on semi-supervised neighbour
Method, it is characterised in that: step 4 specifically includes following content:
Step 4.1 is directed to any video submanifold Ci, calculates the number for belonging to the marked video frame of any class label lj in Ci
Ni;
Step 4.2, using majority vote rule, by the class label l comprising most samples*It is assigned to submanifold Ci;
Step 4.3, the mass center wi, as class label l for calculating video submanifold Ci*Vision code book Wi;
Step 4.4, all vision code books constitute more visual dictionaries.
5. the intelligent video analysis according to claim 4 for propagating study and more visual dictionary models based on semi-supervised neighbour
Method, it is characterised in that: step 5 specifically includes following content:
Video to be checked is kept strategy to extract key video sequence frame by step 5.1 using sampling;
Step 5.2 calculates the OM feature vector O based on sequential metrics;
Step 5.3 calculates OM feature vector O at a distance from vision code book Wi any in more visual dictionaries;
Step 5.4 finds out the corresponding vision code book Wj of minimum range;
Video to be checked is judged to the affiliated submanifold of vision code book Wj by step 5.5.
6. the intelligent video analysis according to claim 5 for propagating study and more visual dictionary models based on semi-supervised neighbour
Method, it is characterised in that: the self-adapting reconstruction learning method of closed loop feedback in step 6 reconstructs and can adapt to the more of new environment
Visual dictionary specifically includes following content:
Step 6.1, the maximum differentiation distance Dmax and threshold value δ of initialization;
Step 6.2 is directed to key frame of video OM feature vector Oi, compares whether Oi is all larger than at a distance from all vision code books
Dmax;
If step 6.3, Oi are all larger than Dmax at a distance from all vision code books, which is judged to UNKNOWN TYPE video;Otherwise,
Jump procedure 5.3 executes;
If step 6.4, UNKNOWN TYPE video number are greater than threshold value δ, 3 are gone to step, construct the more visual words to shake down
Allusion quotation further judges the type of video to be checked, otherwise, terminates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511022492.5A CN105654054B (en) | 2015-12-30 | 2015-12-30 | The intelligent video analysis method of study and more visual dictionary models is propagated based on semi-supervised neighbour |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511022492.5A CN105654054B (en) | 2015-12-30 | 2015-12-30 | The intelligent video analysis method of study and more visual dictionary models is propagated based on semi-supervised neighbour |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105654054A CN105654054A (en) | 2016-06-08 |
CN105654054B true CN105654054B (en) | 2018-12-04 |
Family
ID=56490723
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201511022492.5A Active CN105654054B (en) | 2015-12-30 | 2015-12-30 | The intelligent video analysis method of study and more visual dictionary models is propagated based on semi-supervised neighbour |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105654054B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110475129B (en) | 2018-03-05 | 2021-05-28 | 腾讯科技(深圳)有限公司 | Video processing method, medium, and server |
CN109684506B (en) * | 2018-11-22 | 2023-10-20 | 三六零科技集团有限公司 | Video tagging processing method and device and computing equipment |
CN110390356B (en) * | 2019-07-03 | 2022-03-08 | Oppo广东移动通信有限公司 | Visual dictionary generation method and device and storage medium |
CN113344932B (en) * | 2021-06-01 | 2022-05-03 | 电子科技大学 | Semi-supervised single-target video segmentation method |
CN113515383B (en) * | 2021-07-28 | 2024-02-20 | 中国工商银行股份有限公司 | System resource data distribution method and device |
CN117253196B (en) * | 2023-11-17 | 2024-02-02 | 本溪钢铁(集团)信息自动化有限责任公司 | Video-based security risk monitoring method and device in steel industry |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102855473A (en) * | 2012-08-21 | 2013-01-02 | 中国科学院信息工程研究所 | Image multi-target detecting method based on similarity measurement |
CN103390040A (en) * | 2013-07-17 | 2013-11-13 | 南京邮电大学 | Video copy detection method |
US9098749B2 (en) * | 2013-03-14 | 2015-08-04 | Xerox Corporation | Dictionary design for computationally efficient video anomaly detection via sparse reconstruction techniques |
CN105138991A (en) * | 2015-08-27 | 2015-12-09 | 山东工商学院 | Video emotion identification method based on emotion significant feature integration |
-
2015
- 2015-12-30 CN CN201511022492.5A patent/CN105654054B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102855473A (en) * | 2012-08-21 | 2013-01-02 | 中国科学院信息工程研究所 | Image multi-target detecting method based on similarity measurement |
US9098749B2 (en) * | 2013-03-14 | 2015-08-04 | Xerox Corporation | Dictionary design for computationally efficient video anomaly detection via sparse reconstruction techniques |
CN103390040A (en) * | 2013-07-17 | 2013-11-13 | 南京邮电大学 | Video copy detection method |
CN105138991A (en) * | 2015-08-27 | 2015-12-09 | 山东工商学院 | Video emotion identification method based on emotion significant feature integration |
Also Published As
Publication number | Publication date |
---|---|
CN105654054A (en) | 2016-06-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105654054B (en) | The intelligent video analysis method of study and more visual dictionary models is propagated based on semi-supervised neighbour | |
Shen et al. | Generative adversarial learning towards fast weakly supervised detection | |
CN112446342B (en) | Key frame recognition model training method, recognition method and device | |
CN109743642B (en) | Video abstract generation method based on hierarchical recurrent neural network | |
CN110633708A (en) | Deep network significance detection method based on global model and local optimization | |
Meng et al. | Feature adaptive co-segmentation by complexity awareness | |
CN109583379A (en) | A kind of pedestrian's recognition methods again being aligned network based on selective erasing pedestrian | |
CN112733965B (en) | Label-free image classification method based on small sample learning | |
Wang et al. | Cattle face recognition method based on parameter transfer and deep learning | |
CN108960142B (en) | Pedestrian re-identification method based on global feature loss function | |
CN113761259A (en) | Image processing method and device and computer equipment | |
CN107862680B (en) | Target tracking optimization method based on correlation filter | |
WO2022218396A1 (en) | Image processing method and apparatus, and computer readable storage medium | |
CN112364791B (en) | Pedestrian re-identification method and system based on generation of confrontation network | |
CN109190472A (en) | Combine pedestrian's attribute recognition approach of guidance with attribute based on image | |
Zhang et al. | 3D object retrieval with multi-feature collaboration and bipartite graph matching | |
CN106778714B (en) | LDA face identification method based on nonlinear characteristic and model combination | |
Wang et al. | Spatial weighting for bag-of-features based image retrieval | |
Pratama et al. | Face recognition for presence system by using residual networks-50 architecture | |
Vainstein et al. | Modeling video activity with dynamic phrases and its application to action recognition in tennis videos | |
CN107633527B (en) | Target tracking method and device based on full convolution neural network | |
CN112949658B (en) | Deep learning method with stable performance | |
Qin et al. | Structure-aware feature disentanglement with knowledge transfer for appearance-changing place recognition | |
CN109241315A (en) | A kind of fast face search method based on deep learning | |
CN109902690A (en) | Image recognition technology |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20210701 Address after: Room 690, 6th floor, 999 Changning Road, Changning District, Shanghai 200336 Patentee after: SHANGHAI GUIHE SOFTWARE TECHNOLOGY Co.,Ltd. Address before: Room 801, building 67, 421 Hongcao Road, Xuhui District, Shanghai 200233 Patentee before: SHANGHAI YIBEN INFORMATION TECHNOLOGY Co.,Ltd. |
|
TR01 | Transfer of patent right |