CN103942778A - Fast video key frame extraction method of principal component characteristic curve analysis - Google Patents

Fast video key frame extraction method of principal component characteristic curve analysis Download PDF

Info

Publication number
CN103942778A
CN103942778A CN201410106838.9A CN201410106838A CN103942778A CN 103942778 A CN103942778 A CN 103942778A CN 201410106838 A CN201410106838 A CN 201410106838A CN 103942778 A CN103942778 A CN 103942778A
Authority
CN
China
Prior art keywords
video
low
key frame
frame
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
Application number
CN201410106838.9A
Other languages
Chinese (zh)
Inventor
陈晋音
黄坚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
HANGZHOU XISONG TECHNOLOGY Co Ltd
Original Assignee
HANGZHOU XISONG TECHNOLOGY Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by HANGZHOU XISONG TECHNOLOGY Co Ltd filed Critical HANGZHOU XISONG TECHNOLOGY Co Ltd
Priority to CN201410106838.9A priority Critical patent/CN103942778A/en
Publication of CN103942778A publication Critical patent/CN103942778A/en
Pending legal-status Critical Current

Links

Abstract

The invention discloses a fast video key frame extraction method of the principle characteristic curve analysis. The method includes the steps that firstly, a visual characteristic expression is extracted from video image frames; secondly, PCA dimension reduction processing is performed on a characteristic matrix X composed of all video image frame characteristics, and an image low-dimension characteristic expression Y is obtained; thirdly, each characteristic curve in the Y is searched for a curve local extreme point, and the image low-dimension characteristic expression of the frames wherein the curve local extreme points exist is added into a candidate key frame low-dimension characteristic set T; finally, K mean value clustering is performed on the candidate key frame low-dimension characteristic set, the original video sequence image frame corresponding to the candidate key frame low-dimension characteristic and is closest to the clustering center is used as a final video key frame to be sent back, and therefore extraction of the video key frame is obtained. The fast video key frame extraction method of the principle characteristic curve analysis has the advantages of being small in calculation amount, simple and easy to achieved, and influences of variations of object movement, colors, illumination and the like existing in a video image sequence can be effectively resisted.

Description

A kind of fast video extraction method of key frame of major component feature curve analysis
Technical field
The present invention relates to key frame of video extractive technique, Data Dimensionality Reduction technology and clustering technique, relate in particular to and adopt clustering technique to carry out the method that key frame of video shifts to an earlier date.
Background technology
Along with the develop rapidly of network and multimedia technology, video data presents explosive growth day by day.These video datas are effectively managed and index, just need to be adopted efficient shot detection algorithm and key frame of video extraction algorithm.Involved in the present invention is key frame of video extractive technique, therefore below relevant key frame of video extractive technique is carried out to brief review.
Key frame refers to picture frame most important, the most representative in video lens, and it has comprised camera lens most semantic informations to be expressed.To video extraction key frame, on the one hand can so that user video is browsed fast, on the other hand by the algorithm of originally whole section of video being processed (as the identification of video, target detection etc.), can be transformed into processing accordingly on key frame of video, thus the efficiency of raising algorithm.
Typical key frame of video extraction algorithm, can be divided into following five classes haply:
First kind method is to extract key frame based on the crucial shot boundary of video.Conventionally the key frame of these class methods using the first frame of camera lens and tail frame as camera lens.Comparatively speaking, these class methods are simple, and number of key frames is determined, but practical application effect unstable.In addition, these class methods depend on shot boundary and detect, and in fact the latter itself is also a more unmanageable problem.
Equations of The Second Kind method is to extract key frame by the distance of calculating between consecutive frame.Relatively be typically the method that use curve is simplified.These class methods are comparatively flexible, but easily occur undetected situation, and the key frame extracting also easily comprises similar frame.
The 3rd class methods are the methods based on Video Motion Detection and analysis.Such as adopting optical flow method, Wolf analyzes camera lens in video.The needed calculated amount of these class methods is larger, and efficiency of algorithm is not high, but the key frame extracting relatively meets video motion information.
The 4th class methods are to adopt various clustering methods.These class methods are that video is carried out to cluster according to its visual signature, then by the cluster centre frame of every class, as the key frame of video.The number of key frames that these class methods are extracted is definite, and the weak point of existence is to only have in the time there is obvious Clustering features between frame of video, and when clusters number is more consistent with real data clusters distribution simultaneously, algorithm effect is relatively good.Otherwise, due to the existing restriction of clustering algorithm itself, cause accurately extracting key frame.
The 5th class methods are the thinkings that adopt duty Optimization.These class methods are normally converted into optimization problem of equal value by key frame of video extraction problem, then adopt specific Optimization Solution algorithm to optimize corresponding mathematical model, finally extract the key frame of video according to the result of optimizing.
Method proposed by the invention belongs to the 4th class methods, express by video image being extracted to multiple visual signature, directly image characteristic matrix is done to PCA out, overcome the object of which movement existing in video training, the impact that the variation such as color and illumination brings, then by major component characteristic curve is done to tracing analysis, search curve Local Extremum, obtain candidate's key frame low-dimensional characteristic set, finally on candidate's key frame low-dimensional characteristic set, do K mean cluster, to return as final video key frame apart from the picture frame in the nearest corresponding original video sequence of the low dimensional feature of candidate's key frame of cluster centre, thereby realize the extraction to key frame of video.
Summary of the invention
The object of the invention is to overcome the deficiency of existing key frame of video extraction algorithm, a kind of fast video extraction method of key frame of major component feature curve analysis is provided.
The fast video extraction method of key frame of major component feature curve analysis comprises the steps:
1) for one section of video sequence that comprises n frame video image, extract the SIFT on all images in video sequence, HOG, tetra-kinds of Image Visual Feature of GIST and PHOI, and these four kinds of Image Visual Feature are stitched together successively, obtain the eigenmatrix X=[x that video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, the dimension that wherein d is Image Visual Feature;
2) eigenmatrix X is done to PCA dimension-reduction treatment: first every row in eigenmatrix X are deducted to mean vector ? obtain then right do Eigenvalues Decomposition, m the maximum corresponding proper vector of eigenwert before retaining, by this m proper vector composition low-dimensional Projection Character matrix W=[v 1, v 2..., v m] ∈ R d × m, finally obtain image low-dimensional feature representation Y=W tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
3) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain in P cluster, to return as final video key frame apart from the picture frame in P the nearest corresponding original video sequence of the low dimensional feature of candidate's key frame of cluster centre, thereby realize the extraction to key frame of video.
It is little that this method has calculated amount, is simply easy to the feature that realizes, can effectively resist the impact that the object of which movement, color and the illumination that exist in sequence of video images etc. change, and can realize fast video image key-frame extraction function.
Brief description of the drawings
Fig. 1 is the key-frame extraction result figure of one section of natural land video;
Fig. 2 is the key-frame extraction result figure of one section of Taiji sports video;
Fig. 3 is the key-frame extraction result figure of one section of news report video;
Fig. 4 is the key-frame extraction result figure of one section of daily behavior video.
Embodiment
The fast video extraction method of key frame of major component feature curve analysis comprises the steps:
1) for one section of video sequence that comprises n frame video image, extract the SIFT[1 on all images in video sequence], HOG[2], GIST[3] and PHOI[4] four kinds of Image Visual Feature, and these four kinds of Image Visual Feature are stitched together successively, obtain the eigenmatrix X=[x that video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, the dimension that wherein d is Image Visual Feature;
2) eigenmatrix X is done to PCA dimension-reduction treatment: first every row in eigenmatrix X are deducted to mean vector ? obtain then right do Eigenvalues Decomposition, m the maximum corresponding proper vector of eigenwert before retaining, by this m proper vector composition low-dimensional Projection Character matrix W=[v 1, v 2..., v m] ∈ R d × m, finally obtain image low-dimensional feature representation Y=W tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
3) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain in P cluster, to return as final video key frame apart from the picture frame in P the nearest corresponding original video sequence of the low dimensional feature of candidate's key frame of cluster centre, thereby realize the extraction to key frame of video.
List of references
http://en.wikipedia.org/wiki/Scale-invariant_feature_transform
http://en.wikipedia.org/wiki/Histogram_of_oriented_gradients
Oliva,Aude,and?Antonio?Torralba."Modeling?the?shape?of?the?scene:A?holistic?representation?of?the?spatial?envelope."International?journal?of?computer?vision42.3(2001):145-175.
http://www.vlfeat.org/
Embodiment 1
To one section of natural land video sequence that comprises 11096 frames, according to foregoing method:
1) from video frame images, extract SIFT, HOG, tetra-kinds of Image Visual Feature of GIST and PHOI, amount to the feature representation of 500+576+512+2040=3628 dimension, the Image Visual Feature of above-mentioned Four types is stitched together successively, obtains the eigenmatrix X=[x that this video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, wherein d is 3628 dimensions, n=11096;
2) eigenmatrix X is done to PCA dimension-reduction treatment, obtain the image low-dimensional feature representation Y=W of 500 dimensions tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
3) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain 15 cluster centres, to return as final video key frame apart from the picture frame in 15 nearest corresponding original video sequences of the low dimensional feature of candidate's key frame of cluster centre, result as shown in Figure 1.
Embodiment 2
To one section of Taiji sports video video sequence that comprises 12444 frames, according to foregoing method:
4) from video frame images, extract SIFT, HOG, tetra-kinds of Image Visual Feature of GIST and PHOI, amount to the feature representation of 500+576+512+2040=3628 dimension, the Image Visual Feature of above-mentioned Four types is stitched together successively, obtains the eigenmatrix X=[x that this video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, wherein d is 3628 dimensions, n=12444;
5) eigenmatrix X is done to PCA dimension-reduction treatment, obtain the image low-dimensional feature representation Y=W of 500 dimensions tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
6) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain 20 cluster centres, to return as final video key frame apart from the picture frame in 20 nearest corresponding original video sequences of the low dimensional feature of candidate's key frame of cluster centre, result as shown in Figure 2.
Embodiment 3
To one section of BBC news video sequence that comprises 2340 frames, according to foregoing method:
7) from video frame images, extract SIFT, HOG, tetra-kinds of Image Visual Feature of GIST and PHOI, amount to the feature representation of 500+576+512+2040=3628 dimension, the Image Visual Feature of above-mentioned Four types is stitched together successively, obtains the eigenmatrix X=[x that this video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, wherein d is 3628 dimensions, n=2340;
8) eigenmatrix X is done to PCA dimension-reduction treatment, obtain the image low-dimensional feature representation Y=W of 500 dimensions tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
9) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain 20 cluster centres, to return as final video key frame apart from the picture frame in 20 nearest corresponding original video sequences of the low dimensional feature of candidate's key frame of cluster centre, result as shown in Figure 3.
Embodiment 4
To one section of daily behavior video sequence that comprises 540 frames, according to foregoing method:
10) from video frame images, extract SIFT, HOG, tetra-kinds of Image Visual Feature of GIST and PHOI, amount to the feature representation of 500+576+512+2040=3628 dimension, the Image Visual Feature of above-mentioned Four types is stitched together successively, obtains the eigenmatrix X=[x that this video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, wherein d is 3628 dimensions, n=540;
11) eigenmatrix X is done to PCA dimension-reduction treatment, obtain the image low-dimensional feature representation Y=W of 300 dimensions tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
12) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain 20 cluster centres, to return as final video key frame apart from the picture frame in 20 nearest corresponding original video sequences of the low dimensional feature of candidate's key frame of cluster centre, result as shown in Figure 4.

Claims (1)

1. a fast video extraction method of key frame for major component feature curve analysis, is characterized in that comprising the steps:
1) for one section of video sequence that comprises n frame video image, extract the SIFT on all images in video sequence, HOG, tetra-kinds of Image Visual Feature of GIST and PHOI, and these four kinds of Image Visual Feature are stitched together successively, obtain the eigenmatrix X=[x that video sequence image frame feature forms 1, x 2..., x n] ∈ R d × n, the dimension that wherein d is Image Visual Feature;
2) eigenmatrix X is done to PCA dimension-reduction treatment: first every row in eigenmatrix X are deducted to mean vector ? obtain then right do Eigenvalues Decomposition, m the maximum corresponding proper vector of eigenwert before retaining, by this m proper vector composition low-dimensional Projection Character matrix W=[v 1, v 2..., v m] ∈ R d × m, finally obtain image low-dimensional feature representation Y=W tx=[y 1, y 2..., y n] ∈ R m × n, wherein y ifor the corresponding low-dimensional feature representation of video sequence i frame;
3) regard image low-dimensional feature representation Y the major component characteristic curve of m bar with timing variations as by row, each major component characteristic curve is analyzed, search characteristics curve Local Extremum, the low-dimensional feature representation of the corresponding video frame image of characteristic curve Local Extremum is joined in candidate's key frame low-dimensional characteristic set T, obtain candidate's key frame low-dimensional feature set T={y i| y i∈ set (Y) ∧ y ithe low-dimensional feature representation of the corresponding frame of video of characteristic curve Local Extremum }, wherein set (Y)={ y 1, y 2..., y nit is the corresponding set expression of Y;
4) candidate's key frame low-dimensional characteristic set T is done to K mean cluster, obtain in P cluster, to return as final video key frame apart from the picture frame in P the nearest corresponding original video sequence of the low dimensional feature of candidate's key frame of cluster centre, thereby realize the extraction to key frame of video.
CN201410106838.9A 2014-03-20 2014-03-20 Fast video key frame extraction method of principal component characteristic curve analysis Pending CN103942778A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410106838.9A CN103942778A (en) 2014-03-20 2014-03-20 Fast video key frame extraction method of principal component characteristic curve analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410106838.9A CN103942778A (en) 2014-03-20 2014-03-20 Fast video key frame extraction method of principal component characteristic curve analysis

Publications (1)

Publication Number Publication Date
CN103942778A true CN103942778A (en) 2014-07-23

Family

ID=51190429

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410106838.9A Pending CN103942778A (en) 2014-03-20 2014-03-20 Fast video key frame extraction method of principal component characteristic curve analysis

Country Status (1)

Country Link
CN (1) CN103942778A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108121973A (en) * 2017-12-25 2018-06-05 江苏易乐网络科技有限公司 Key frame extraction method of motion capture data based on principal component analysis
CN108537732A (en) * 2018-04-10 2018-09-14 福州大学 Fast image splicing method based on PCA-SIFT
CN110427825A (en) * 2019-07-01 2019-11-08 上海宝钢工业技术服务有限公司 The video flame recognition methods merged based on key frame with quick support vector machines
CN110674347A (en) * 2019-09-02 2020-01-10 南京邮电大学 Visual shielding double-layer AP video abstract generation method
CN112396551A (en) * 2019-08-16 2021-02-23 阿里巴巴集团控股有限公司 Watermark embedding method and device
CN113257364A (en) * 2021-05-26 2021-08-13 南开大学 Single cell transcriptome sequencing data clustering method and system based on multi-objective evolution
CN113261987A (en) * 2021-03-25 2021-08-17 聚融医疗科技(杭州)有限公司 Three-dimensional ultrasonic imaging method and system based on moving target

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120148157A1 (en) * 2010-12-10 2012-06-14 Mrityunjay Kumar Video key-frame extraction using bi-level sparsity
CN103218824A (en) * 2012-12-24 2013-07-24 大连大学 Motion key frame extracting method based on distance curve amplitudes

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120148157A1 (en) * 2010-12-10 2012-06-14 Mrityunjay Kumar Video key-frame extraction using bi-level sparsity
CN103218824A (en) * 2012-12-24 2013-07-24 大连大学 Motion key frame extracting method based on distance curve amplitudes

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
MJ PARK等: "《Example-based motion cloning》", 《COMPUTER ANIMATION AND VIRTUAL WORLDS》 *
ZHANG QIANG等: "《Motion Key-frames extraction based on amplitude of distance characteristic curve》", 《INTERNATIONAL JOURNAL OF COMPUTATIONAL INTELLIGENCE SYSTEMS》 *
王深: "《基于聚类算法的多特征融合关键帧提取技术研究》", 《万方学位论文》 *
许文竹等: "《结合主成分分析和聚类的关键帧提取》", 《计算机工程与应用》 *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108121973A (en) * 2017-12-25 2018-06-05 江苏易乐网络科技有限公司 Key frame extraction method of motion capture data based on principal component analysis
CN108537732A (en) * 2018-04-10 2018-09-14 福州大学 Fast image splicing method based on PCA-SIFT
CN108537732B (en) * 2018-04-10 2021-11-02 福州大学 PCA-SIFT-based rapid image splicing method
CN110427825A (en) * 2019-07-01 2019-11-08 上海宝钢工业技术服务有限公司 The video flame recognition methods merged based on key frame with quick support vector machines
CN110427825B (en) * 2019-07-01 2023-05-12 上海宝钢工业技术服务有限公司 Video flame identification method based on fusion of key frame and fast support vector machine
CN112396551A (en) * 2019-08-16 2021-02-23 阿里巴巴集团控股有限公司 Watermark embedding method and device
CN110674347A (en) * 2019-09-02 2020-01-10 南京邮电大学 Visual shielding double-layer AP video abstract generation method
CN110674347B (en) * 2019-09-02 2022-04-01 南京邮电大学 Visual shielding double-layer AP video abstract generation method
CN113261987A (en) * 2021-03-25 2021-08-17 聚融医疗科技(杭州)有限公司 Three-dimensional ultrasonic imaging method and system based on moving target
CN113257364A (en) * 2021-05-26 2021-08-13 南开大学 Single cell transcriptome sequencing data clustering method and system based on multi-objective evolution
CN113257364B (en) * 2021-05-26 2022-07-12 南开大学 Single cell transcriptome sequencing data clustering method and system based on multi-objective evolution

Similar Documents

Publication Publication Date Title
Deng et al. Image aesthetic assessment: An experimental survey
Radenović et al. Revisiting oxford and paris: Large-scale image retrieval benchmarking
Uijlings et al. Video classification with densely extracted hog/hof/mbh features: an evaluation of the accuracy/computational efficiency trade-off
CN103942778A (en) Fast video key frame extraction method of principal component characteristic curve analysis
Soomro et al. Action recognition in realistic sports videos
WO2021238062A1 (en) Vehicle tracking method and apparatus, and electronic device
Ren et al. Region-based saliency detection and its application in object recognition
Cech et al. Efficient sequential correspondence selection by cosegmentation
US8983192B2 (en) High-confidence labeling of video volumes in a video sharing service
Rahmani et al. Discriminative human action classification using locality-constrained linear coding
Xiaofeng et al. Discriminatively trained sparse code gradients for contour detection
US8942469B2 (en) Method for classification of videos
CN110866896B (en) Image saliency target detection method based on k-means and level set super-pixel segmentation
CN103988232A (en) IMAGE MATCHING by USING MOTION MANIFOLDS
EP3008632A1 (en) Color sketch image searching
CN111667005B (en) Human interactive system adopting RGBD visual sensing
Wu et al. Scene text detection using adaptive color reduction, adjacent character model and hybrid verification strategy
CN111832568A (en) License plate recognition method, and training method and device of license plate recognition model
Del Pero et al. Behavior discovery and alignment of articulated object classes from unstructured video
Wang et al. Simultaneously discovering and localizing common objects in wild images
Zhou et al. Semantic image segmentation using low-level features and contextual cues
Thyagharajan et al. Prevalent color extraction and indexing
CN109427068A (en) Deep learning based on super-pixel automatic marking is divided into segmentation method
Jiang et al. Object recognition and tracking for indoor robots using an RGB-D sensor
Zhou et al. Modeling perspective effects in photographic composition

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20140723

RJ01 Rejection of invention patent application after publication