CN109195026A - Video abstraction generating method and system - Google Patents
Video abstraction generating method and system Download PDFInfo
- Publication number
- CN109195026A CN109195026A CN201811195007.8A CN201811195007A CN109195026A CN 109195026 A CN109195026 A CN 109195026A CN 201811195007 A CN201811195007 A CN 201811195007A CN 109195026 A CN109195026 A CN 109195026A
- Authority
- CN
- China
- Prior art keywords
- video
- frame
- similarity
- video frame
- characteristic information
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/80—Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
- H04N21/85—Assembly of content; Generation of multimedia applications
- H04N21/854—Content authoring
- H04N21/8549—Creating video summaries, e.g. movie trailer
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computer Security & Cryptography (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of video abstraction generating method and systems, wherein video abstraction generating method includes the method according to dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；Similarity is calculated according to the characteristic information；Adaptive adjustment is carried out to video according to similarity to be segmented video；According to the similarity in the section of the video of segmentation and between section, it is adaptively adjusted similar frame discrimination standard, and similar video frame is merged, generates final video frequency abstract.By using the method for dictionary learning and rarefaction representation, standard, keeps video abstraction generating method more adaptable with being adaptively adjusted video segmentation.
Description
Technical field
The present invention relates to technical field of video processing, and in particular to a kind of video abstraction generating method and system.
Background technique
In recent years, with universal and network technology the development of digital photographing apparatus, it is raw that video is increasingly becoming record people
A kind of important form lived and linked up, in order to save time it is desirable to extract the important content of video,
To quickly understand the key content of video, video frequency abstract is exactly a kind of technology that can satisfy this demand, video frequency abstract algorithm
The importance of its various pieces can be assessed according to video content, and prior extracting section is come out and constitutes view
Frequency is made a summary.However, video type is more, and content is complex, proposes higher want to the design of video frequency abstract algorithm
It asks, and that there are performances is not good enough for the generation method of existing video frequency abstract, and this method is not strong to the universality of different scenes asks
Topic.
Summary of the invention
The object of the present invention is to provide video abstraction generating method and systems.
In order to solve the above-mentioned technical problems, the present invention provides a kind of video abstraction generating methods to include:
According to the method for dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；
Similarity is calculated according to the characteristic information；
Adaptive adjustment is carried out to video according to similarity to be segmented video；
According to the similarity in the section of the video of segmentation and between section, similar video frame is merged, is generated final
Video frequency abstract.
Further, the method for the characteristic information to video extraction video frame includes:
Pretreatment and feature extraction are carried out to the content of video, in the form of representation of video shot is become video frame；
Video frame is carried out down-sampled；
Each video frame is read in, characteristic information is extracted.
Further, the characteristic information includes SIFT feature and HSV feature.
Further, the method for calculating similarity according to the characteristic information includes:
After video extraction characteristic information, each video-frequency band is expressed as a matrix；
Its reconstructed error on corresponding dictionary is defined as follows: | | Xi-DAi||2 F, wherein D indicates dictionary, AiIndicate video
The corresponding reconstruction coefficients of frame, XiIt is the SIFT feature of video extraction；
Judge current video and video before in terms of content according to the variation of the reconstructed error and the reconstruction coefficients
Difference, according to encoder matrix and spatial pyramid, encoder matrix, which is converted to one, multi-scale information vector；It measures described more
Tri- frame of dimensional information vector sum n-1, n-2, n-3 corresponds to the distance of vector, and takes its average value as present frame and previous frame
Similarity；Wherein n indicates present frame.
Further, described that video progress is adaptively adjusted by the method being segmented to video includes: according to similarity
According to the similarity of interframe at video lens fragment position and in video mirror head section, the reasonable threshold of video segmentation is calculated
Value, is segmented video lens；
According to segmentation result, the video frame of each video-frequency band is extracted, and measures the similarity between video frame, to segmentation
As a result it is adaptively adjusted.
Further, the reasonable threshold value for calculating video segmentation, the method being segmented to video lens include:
Step S1 extracts data and feature, each picture frame is expressed as a three-dimensional feature vector, according to consecutive frame
Between similarity numerical curve, extract local minimum, extract the data of the minimum position；
Step S2 carries out initial classification, the nearest neighbor classifier of two classification of building, according to the minimum position to data
The data set find out the smallest data of similarity numerical value, using its corresponding feature vector as the classification of video shot boundary classification
Center；A maximum value is looked for again, it is right using the classifier using its feature vector as the class center for not being boundary classification
Remaining data is classified；
Step S3 reselects positive negative sample, is updated to classifier, and data are carried out with new classification；
Step S4 repeats step S3, until reaching termination condition, i.e. algorithm reaches maximum cycle or data
Classification results no longer change.
Further, the similarity in the section according to the video of segmentation and between section, is adaptively adjusted similar frame and sentences
Other standard, and similar video frame is merged, the method for generating final video frequency abstract includes:
Step A1: the video frame in each segmenting video is extracted according to segmentation result, and measures the phase between each video frame
Like degree；
Step A2 detects that transition in video or content change time are less than the camera lens of preset value according to similarity, and
Carry out delete processing；
Step A3, video frame and corresponding video-frequency band after merging treatment, and generate final video frequency abstract.
Further, the video abstraction generating method further include:
The numerical value of Fscore according to final video frequency abstract evaluates the quality of abstract, wherein described
Fscore is that the measurement index accurate rate (precision) combined and recall rate (recall) are calculated, and formula is as follows:
Wherein: prescision=Nmatch/NAS；Recall=Nmatch/NUS
Wherein NmatchIt is that the video video frame that algorithm generates matches the artificial quantity for extracting video frame: NASAnd NUSRespectively
Indicate that algorithm is the quantity for generating video frame and manually selected video frame.
The present invention also provides a kind of video frequency abstracts to generate system, comprising:
Characteristic extracting module is suitable for the method according to dictionary learning and rarefaction representation, believes the feature of video extraction video frame
Breath；
Similarity calculation module is suitable for calculating similarity according to the characteristic information；
Segmentation module is suitable for carrying out adaptive adjustment to video according to similarity being segmented video；
Summarization generation module is suitable for being carried out similar video frame according to the similarity in the section of the video of segmentation and between section
Merge, generates final video frequency abstract.
Further, the video frequency abstract generates system further include:
Quality assessment module is suitable for the numerical value of the Fscore according to final video frequency abstract to comment the quality of abstract
Valence, the Fscore are that the measurement index accurate rate (precision) combined and recall rate (recall) are calculated, and guarantee essence
True rate and recall rate are all higher.However in actual experiment, accurate rate and recall rate are difficult to get both, using the numerical value of the Fscore
It replaces, formula is as follows:
Wherein: prescision=Nmatch/NAS；Recall=Nmatch/NUS
Wherein NmatchIt is that the video video frame that algorithm generates matches the artificial quantity for extracting video frame: NASAnd NUSRespectively
Indicate that algorithm is the quantity for generating video frame and manually selected video frame.
The invention has the advantages that the present invention provides a kind of video abstraction generating method and systems, wherein video is plucked
Wanting generation method includes the method according to dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；According to institute
It states characteristic information and calculates similarity；Adaptive adjustment is carried out to video according to similarity to be segmented video；According to segmentation
Similarity in the section of video and between section, is adaptively adjusted similar frame discrimination standard, and similar video frame is merged,
Generate final video frequency abstract.By using the method for dictionary learning and rarefaction representation, it is adaptively adjusted video segmentation terrestrial reference
Standard keeps video abstraction generating method more adaptable.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 shows the flow diagram of video abstraction generating method provided by the embodiment of the present invention.
Fig. 2 is the sub-step flow chart of step S130 in Fig. 1.
Fig. 3 is the sub-step flow chart of step S131 in Fig. 2.
Fig. 4 shows the functional block diagram that video frequency abstract provided by the embodiment of the present invention generates system.
Specific embodiment
In conjunction with the accompanying drawings, the present invention is further explained in detail.These attached drawings are simplified schematic diagram, only with
Illustration illustrates basic structure of the invention, therefore it only shows the composition relevant to the invention.
Embodiment is as shown in Figure 1, the embodiment of the invention provides a kind of video abstraction generating methods.Video frequency abstract generation side
Method includes the next steps:
S110: according to the method for dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame.
By using the method for dictionary learning and rarefaction representation, the characteristic information of adaptive extraction video frame is realized
Video abstraction generating method is more adaptable.Wherein, following methods are used to the characteristic information of video extraction video frame:
Pretreatment and feature extraction are carried out to the content of video, in the form of representation of video shot is become video frame；
It is down-sampled to video frame progress, it is in the present embodiment, down-sampled that the quantity of video frame is reduced to original 1/10,
In other embodiments, down-sampled quantity can sets itself, it is within the scope of the present invention.By down-sampled by video
The quantity of frame is reduced, and the data processing quantity of processor and the occupancy of memory are reduced.
Each video frame is read in, characteristic information is extracted.In the present embodiment, characteristic information includes SIFT spy
It seeks peace HSV feature.
S120: similarity is calculated according to the characteristic information.
The characteristic information of adaptive extraction video frame is adaptively adjusted video according to the similarity for calculating video frame
Piecewise standard, it is more adaptable to realize video abstraction generating method.Wherein, step S120 the following steps are included:
After video extraction characteristic information, each video-frequency band is expressed as a matrix；
Its reconstructed error on corresponding dictionary is defined as follows: | | Xi-DAi||2 F, wherein D indicates dictionary, AiIndicate video
The corresponding reconstruction coefficients of frame, XiIt is the SIFT feature of video extraction；
Judge current video and video before in terms of content according to the variation of the reconstructed error and the reconstruction coefficients
Difference, according to encoder matrix and spatial pyramid, encoder matrix, which is converted to one, multi-scale information vector；It measures described more
Tri- frame of dimensional information vector sum n-1, n-2, n-3 corresponds to the distance of vector, and takes its average value as present frame and previous frame
Similarity；Wherein n indicates present frame.
S130: adaptive adjustment is carried out to video according to similarity, video is segmented.
Please refer to 2, wherein adaptive adjustment is carried out to video according to similarity, segmentation is carried out using following step to video
It is rapid:
S131: according to the similarity of interframe at video lens fragment position and in video mirror head section, the conjunction of video segmentation is calculated
Threshold value is managed, video lens are segmented, referring to Fig. 3, in the present embodiment, the method packet that video lens are segmented
It includes:
Step S1: data and feature are extracted, each picture frame is expressed as a three-dimensional feature vector, according to consecutive frame
Between similarity numerical curve, extract local minimum, extract the data of the minimum position；
Step S2: data are carried out with initial classification, the nearest neighbor classifier of two classification of building, according to the minimum position
The data set find out the smallest data of similarity numerical value, using its corresponding feature vector as the classification of video shot boundary classification
Center；A maximum value is looked for again, it is right using the classifier using its feature vector as the class center for not being boundary classification
Remaining data is classified；
Step S3: reselecting positive negative sample, be updated to classifier, and data are carried out with new classification；
Step S4: repeating step S3, until reaching termination condition, i.e. algorithm reaches maximum cycle or data
Classification results no longer change.
S132: according to segmentation result, extracting the video frame of each video-frequency band, and measure the similarity between video frame,
Segmentation result is adaptively adjusted.
S140: according to the similarity in the section of the video of segmentation and between section, similar frame can also be adaptively adjusted and differentiated
Then standard again merges similar video frame, generate final video frequency abstract.
By the way that adaptive video frame extraction algorithm, including the deletion of filtering camera lens section and similar video frame two rings of merging are arranged
Section, by adaptive video segmentation and video frame extraction algorithm, so that the video frequency abstract quality generated greatly improves.Wherein,
Step S140 the following steps are included:
Step A1: extracting the video frame in each segmenting video according to segmentation result, and measures similar between video frame
Degree；
Step A2: detecting that transition in video or content change time are less than the camera lens of preset value according to similarity, and
Delete processing is carried out, in the present embodiment, the video frame for being less than the camera lens of preset value for transition or content change time carries out
Delete processing, to improve the precision of sample.
Step A3: video frame and corresponding video-frequency band after merging treatment, and generate final video frequency abstract.
In the present embodiment, the similar frame discrimination standard, that is, first according to adaptive video segmentation result, extract every
The video frame of a video mirror head section, and measure the similarity of video interframe.Then, it according to these similarities, detects in video
Transition or the too fast camera lens section of content change, and handled；Finally merge remaining video frame and corresponding video-frequency band,
And generate new video frame.
In the present embodiment, the video abstraction generating method further include:
Step S150: the numerical value of the Fscore according to final video frequency abstract evaluates the quality of abstract.
Fscore be combine measurement index accurate rate (precision) and recall rate (recall) calculated, guarantee accurate rate with
Recall rate is all higher.However in actual experiment, accurate rate and recall rate are difficult to get both, using the numerical value of the Fscore come generation
It replaces, formula is as follows:
Wherein: prescision=Nmatch/NAS；Recall=Nmatch/NUS, wherein NmatchIt is the video view that algorithm generates
Frequency frame matches the artificial quantity for extracting video frame: NASAnd NUSRespectively indicating algorithm is to generate video frame and manually selected view
The quantity of frequency frame.The quality of abstract is evaluated by the numerical value of the Fscore to abstract, convenient for being generated to video frequency abstract
The value of parameter inside method is adjusted.
Referring to Fig. 4, the present invention also provides a kind of video frequency abstracts to generate system.It includes feature that video frequency abstract, which generates system,
Extraction module is suitable for the method according to dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；Similarity meter
Module is calculated to be suitable for calculating similarity according to the characteristic information；Segmentation module is suitable for adaptively adjusting video according to similarity
It is whole that video is segmented；Summarization generation module is suitable for according to the similarity in the section of the video of segmentation and between section, can also be certainly
Similar frame discrimination standard is adaptively adjusted, then similar video frame is merged, generates final video frequency abstract.By using
The method of dictionary learning and rarefaction representation, standard, has video abstraction generating method more with being adaptively adjusted video segmentation
There is adaptability.
In the present embodiment, the video frequency abstract generates system further include: quality assessment module is suitable for according to final view
The numerical value of the Fscore of frequency abstract evaluates the quality of abstract.Quality assessment module is suitable for according to final video frequency abstract
The numerical value of Fscore the quality of abstract is evaluated, the Fscore is the measurement index accurate rate combined
(precision) it is calculated with recall rate (recall), guarantees that accurate rate and recall rate are all higher.However in actual experiment,
Accurate rate and recall rate are difficult to get both, and are replaced using the numerical value of the Fscore, formula is as follows:
Wherein: prescision=Nmatch/NAS；Recall=Nmatch/NUS, wherein NmatchIt is the video that algorithm generates
Video frame matches the artificial quantity for extracting video frame: NASAnd NUSRespectively indicating algorithm is to generate video frame and manually select
The quantity of video frame.The quality of abstract is evaluated by the numerical value of the Fscore to abstract, convenient for raw to video frequency abstract
It is adjusted at the value of the parameter inside method.
In conclusion the present invention provides a kind of video abstraction generating method and systems, wherein video abstraction generating method
Including the method according to dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；According to the characteristic information
Calculate similarity；Adaptive adjustment is carried out to video according to similarity to be segmented video；According in the section of the video of segmentation
And the similarity between section, it is adaptively adjusted similar frame discrimination standard, and similar video frame is merged, generated final
Video frequency abstract.By using the method for dictionary learning and rarefaction representation, standard, plucks video with being adaptively adjusted video segmentation
Want generation method more adaptable.
Taking the above-mentioned ideal embodiment according to the present invention as inspiration, through the above description, relevant staff is complete
Various changes and amendments can be carried out without departing from the scope of the technological thought of the present invention' entirely.The technology of this invention
Property range is not limited to the contents of the specification, it is necessary to which the technical scope thereof is determined according to the scope of the claim.
Claims (10)
1. a kind of video abstraction generating method characterized by comprising
According to the method for dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；
Similarity is calculated according to the characteristic information；
Adaptive adjustment is carried out to video according to similarity to be segmented video；
According to the similarity in the section of the video of segmentation and between section, similar video frame is merged, generates final video
Abstract.
2. video abstraction generating method as described in claim 1, which is characterized in that
The method of the characteristic information to video extraction video frame includes:
Pretreatment and feature extraction are carried out to the content of video, in the form of representation of video shot is become video frame；
Video frame is carried out down-sampled；
Each video frame is read in, characteristic information is extracted.
3. video abstraction generating method as claimed in claim 2, which is characterized in that the characteristic information include SIFT feature and
HSV feature.
4. video abstraction generating method as claimed in claim 3, which is characterized in that
It is described according to the characteristic information calculate similarity method include:
After video extraction characteristic information, each video-frequency band is expressed as a matrix；
Its reconstructed error on corresponding dictionary is defined as follows: | | Xi-DAi||2 F, wherein D indicates dictionary, AiIndicate video frame pair
The reconstruction coefficients answered, XiIt is the SIFT feature of video extraction；
Current video and the difference of video in terms of content before are judged according to the variation of the reconstructed error and the reconstruction coefficients,
According to encoder matrix and spatial pyramid, encoder matrix, which is converted to one, multi-scale information vector；It measures described multiple dimensioned
Information vector and n-1, tri- frame of n-2, n-3 correspond to the distance of vector, and take its average value as the similar of present frame and previous frame
Degree；Wherein n indicates present frame.
5. video abstraction generating method as claimed in claim 4, which is characterized in that
The foundation similarity adaptively adjusts the method being segmented to video to video progress
According to the similarity of interframe at video lens fragment position and in video mirror head section, the reasonable threshold value of video segmentation is calculated, it is right
Video lens are segmented；
According to segmentation result, the video frame of each video-frequency band is extracted, and measures the similarity between video frame, to segmentation result
Adaptively adjusted.
6. video abstraction generating method as claimed in claim 5, which is characterized in that
The reasonable threshold value for calculating video segmentation, the method being segmented to video lens include:
Step S1 extracts data and feature, each picture frame is expressed as a three-dimensional feature vector, according between consecutive frame
Similarity numerical curve, extract local minimum, extract the data of the minimum position；
Step S2 carries out initial classification, the nearest neighbor classifier of two classification of building, according to the minimum position to data
Data find out the smallest data of similarity numerical value, using its corresponding feature vector as in the classification of video shot boundary classification
The heart；A maximum value is looked for again, using its feature vector as the class center for not being boundary classification, using the classifier, to surplus
Remainder is according to classifying；
Step S3 reselects positive negative sample, is updated to classifier, and data are carried out with new classification；
Step S4 repeats step S3, until reach termination condition, i.e. algorithm point that reaches maximum cycle or data
Class result no longer changes.
7. video abstraction generating method as described in claim 1, which is characterized in that
Similarity in the section according to the video of segmentation and between section, similar video frame is merged, and is generated final
The method of video frequency abstract includes:
Step A1 extracts the video frame in each segmenting video according to segmentation result, and measures the similarity between each video frame；
Step A2 detects that transition or content change time in video are less than the camera lens of preset value according to similarity, and carries out
Delete processing；
Step A3, video frame and corresponding video-frequency band after merging treatment, and generate final video frequency abstract.
8. video abstraction generating method as described in claim 1, which is characterized in that the video abstraction generating method also wraps
Include: the numerical value of the Fscore according to final video frequency abstract evaluates the quality of abstract, wherein the Fscore is knot
The measurement index accurate rate (precision) and recall rate (recall) of conjunction are calculated, and formula is as follows:
Wherein: prescision=Nmatch/NAS；Recall=Nmatch/NUS
Wherein NmatchIt is that the video video frame that algorithm generates matches the artificial quantity for extracting video frame: NASAnd NUSIt respectively indicates
Algorithm is the quantity for generating video frame and manually selected video frame.
9. a kind of video frequency abstract generates system characterized by comprising
Characteristic extracting module is suitable for the method according to dictionary learning and rarefaction representation, to the characteristic information of video extraction video frame；
Similarity calculation module is suitable for calculating similarity according to the characteristic information；
Segmentation module is suitable for carrying out adaptive adjustment to video according to similarity being segmented video；
Summarization generation module is suitable for being closed similar video frame according to the similarity in the section of the video of segmentation and between section
And generate final video frequency abstract.
10. video frequency abstract as claimed in claim 9 generates system, which is characterized in that the video frequency abstract generates system and also wraps
It includes:
Quality assessment module is suitable for the numerical value of the Fscore according to final video frequency abstract to evaluate the quality of abstract.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811195007.8A CN109195026A (en) | 2018-10-15 | 2018-10-15 | Video abstraction generating method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811195007.8A CN109195026A (en) | 2018-10-15 | 2018-10-15 | Video abstraction generating method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109195026A true CN109195026A (en) | 2019-01-11 |
Family
ID=64944461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811195007.8A Pending CN109195026A (en) | 2018-10-15 | 2018-10-15 | Video abstraction generating method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109195026A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110267040A (en) * | 2019-06-27 | 2019-09-20 | 国网山东省电力公司建设公司 | A kind of method for compressing image based on video flow detection |
| CN111586473A (en) * | 2020-05-20 | 2020-08-25 | 北京字节跳动网络技术有限公司 | Video clipping method, device, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106034264A (en) * | 2015-03-11 | 2016-10-19 | 中国科学院西安光学精密机械研究所 | Coordination-model-based method for obtaining video abstract |
| CN106056627A (en) * | 2016-05-30 | 2016-10-26 | 河海大学 | Robustness object tracking method based on local identification sparse representation |
-
2018
- 2018-10-15 CN CN201811195007.8A patent/CN109195026A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106034264A (en) * | 2015-03-11 | 2016-10-19 | 中国科学院西安光学精密机械研究所 | Coordination-model-based method for obtaining video abstract |
| CN106056627A (en) * | 2016-05-30 | 2016-10-26 | 河海大学 | Robustness object tracking method based on local identification sparse representation |
Non-Patent Citations (1)
| Title |
|---|
| 李佳桐: "自适应视频摘要算法研究", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110267040A (en) * | 2019-06-27 | 2019-09-20 | 国网山东省电力公司建设公司 | A kind of method for compressing image based on video flow detection |
| CN111586473A (en) * | 2020-05-20 | 2020-08-25 | 北京字节跳动网络技术有限公司 | Video clipping method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111104898B (en) | Image scene classification method and device based on target semantics and attention mechanism | |
| CN108830855B (en) | Full convolution network semantic segmentation method based on multi-scale low-level feature fusion | |
| CN109151501A (en) | A kind of video key frame extracting method, device, terminal device and storage medium | |
| Cong et al. | Towards scalable summarization of consumer videos via sparse dictionary selection | |
| CN103593464B (en) | Video fingerprint detecting and video sequence matching method and system based on visual features | |
| US7869657B2 (en) | System and method for comparing images using an edit distance | |
| Johnson et al. | Sparse coding for alpha matting | |
| CN102236796B (en) | Method and system for sorting defective contents of digital video | |
| WO2002077909A1 (en) | Video segmentation using statistical pixel modeling | |
| CN108921130A (en) | Video key frame extracting method based on salient region | |
| JP2004199669A (en) | Face detection | |
| JP2004192637A (en) | Face detection | |
| CN113112519B (en) | Key frame screening method based on interested target distribution | |
| CN111768388A (en) | Product surface defect detection method and system based on positive sample reference | |
| WO2023066173A1 (en) | Image processing method and apparatus, and storage medium and electronic device | |
| CN109195026A (en) | Video abstraction generating method and system | |
| CN116030396B (en) | Accurate segmentation method for video structured extraction | |
| WO2023217046A1 (en) | Image processing method and apparatus, nonvolatile readable storage medium and electronic device | |
| CN110188625B (en) | Video fine structuring method based on multi-feature fusion | |
| Zhao et al. | Detecting deepfake video by learning two-level features with two-stream convolutional neural network | |
| KR20210011707A (en) | A CNN-based Scene classifier with attention model for scene recognition in video | |
| Zhang | [Retracted] Sports Action Recognition Based on Particle Swarm Optimization Neural Networks | |
| CN105893967B (en) | Human behavior classification detection method and system based on time sequence retention space-time characteristics | |
| Chen et al. | An adaptive noise removal tool for iot image processing under influence of weather conditions | |
| Liu et al. | Key frame extraction based on improved frame blocks features and second extraction |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190111 |
|
| RJ01 | Rejection of invention patent application after publication |