CN102098449B - A kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation - Google Patents
A kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation Download PDFInfo
- Publication number
- CN102098449B CN102098449B CN201010574074.8A CN201010574074A CN102098449B CN 102098449 B CN102098449 B CN 102098449B CN 201010574074 A CN201010574074 A CN 201010574074A CN 102098449 B CN102098449 B CN 102098449B
- Authority
- CN
- China
- Prior art keywords
- subgraph
- camera lens
- mark
- programme
- program
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to Video processing and mode identification technology, it is proposed that a kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation.At present, the inside division of TV programme has urgent needs；It is good structural that the time discontinuity of program sign makes program have.The method of the invention mainly comprises the steps that program video shot segmentation, each camera lens are extracted key frame and the subgraph of program sign region by (1)；(2) extract the characteristic vector of subgraph, use the SVM classifier for program sign to classify；(3) statistical mark classification results, demarcates the flag property of each camera lens；(4) the flag property shear point at adjacent camera lens splits video.Key frame, during Mark Detection, is only processed, thus improves the efficiency of method by the present invention；It addition, the main application of the present invention is the TV programme that program sign has discontinuity, to program content type no requirement (NR), enhance the universality of method application.
Description
Technical field
The invention belongs to Video processing and mode identification technology, be specifically related to a kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation.
Background technology
At present, radio and television are all producing the video of magnanimity every day, and give electric program menu.Widely available along with Web TV and DTV, more preferably views and admires impression to provide, and many TV programme attempt, by internal paragraph segmentation, providing program-internal rating and instructing.Meanwhile, the inside division of program is also the premise of further content analysis and retrieval.In the face of the video of magnanimity, artificial mark segmentation can not meet timeliness requirement, and what machine completed is partitioned into urgent needs automatically.Video structure analyzing refers to that video flowing carries out shot segmentation, key-frame extraction and scene cut etc. to be processed, thus obtains the structured message of video.Scene cut is concentrated mainly on scene clustering, repeats Video Detection, and shot similarity ratio is on counterpart method, the most more complicated.Currently, increasing TV programme pay close attention to intellectual property when using station symbol or the own mark of program: at the video paragraph of the non-own property right of program-internal, such as advertisement, the vidclip etc. quoted, will not load these marks；And use the video paragraph of mark, it is common that teaser or tail, interview part, or other fragments recorded by this program oneself.It is the strongest structural that mark discontinuity in time series makes TV programme have, and the inside division for TV programme provides foundation.
Summary of the invention
For a certain specific television program, its station symbol or program sign, referred to collectively below as mark, has temporal discontinuity, and the present invention provides a kind of to this kind of TV programme inside division method, reaches segmentation effect quickly and accurately.
The key step of TV programme automatic inside segmentation method of the present invention is as follows:
Step one, utilizes a kind of existing shot segmentation technique that television program video is carried out shot segmentation, it is thus achieved that the shot sequence information of Time Continuous；
Step 2, takes 5 frame key frames, and extracts the subgraph of the rectangular area of special sign position in all key frames each camera lens temporally average mode；
Step 3, extracts the image feature vector of all subgraphs of training set, and the sub-pictures containing mark is positive sample, and the sub-pictures without mark is negative sample, and training obtains SVM classifier；
Step 4, to this program video to be split, all subgraphs are obtained through step (1) and (2), extract the image feature vector identical with step (3), the SVM classifier obtained by step (3) is classified, and obtains the classification results of each subgraph；
Step 5, labelling camera lens flag property, if at least 3 frame subgraphs are judged as existing mark in camera lens, then this camera lens of labelling is mark camera lens, is otherwise labeled as non-mark camera lens；
Step 6, program video inside division, video has the adjacent shot boundary of unlike signal attribute as cut-point, Video segmentation is become paragraph.
Accompanying drawing explanation
Fig. 1 is television program structure exemplary plot of the present invention.
Fig. 2 is the basic flow sheet of the method for the invention.
Detailed description of the invention
As shown in Fig. 2 flow chart, the method for the invention comprises two stages: off-line training grader and online treatment video to be split.Two stages common step is shot segmentation, extracts 5 frame key frames and specific region subgraph thereof.It it is below method detailed description of the invention.
(1) shot segmentation step is to utilize existing a kind of shot segmentation algorithm, as based on rectangular histogram, based on moving and for the algorithm compressing video, specific television program video slicing becoming the shot sequence of Time Continuous.
(2) each camera lens is temporally divided into 6 sections, takes 5 two field pictures of adjacent segment as key frame；For these TV programme, known special sign determining the rectangular area at its place, mark is surrounded by this rectangle just completely, and rectangular coordinates is (x, y, w, h), wherein x, y is respectively the transverse and longitudinal coordinate of rectangle upper left angle point, and w, h are respectively width and the height of rectangle；To this rectangle of all key-frame extraction, referred to as subgraph.
(3) three kinds of image feature vectors of all subgraphs are extracted: HSV space color histogram, edge gradient rectangular histogram, SIFT feature point rectangular histogram based on word bag model；Then three kinds of features are connected, form last image feature vector.Specific features extracting method is as follows:
1. color histogram extracts
Subgraph is extracted hsv color statistic histogram, and wherein H space is divided into 8 intervals, and S space is divided into 3 intervals, and V space is divided into 3 spaces, by rectangular histogram normalization, forms the characteristic vector of 72 dimensions；
2. edge gradient rectangular histogram is extracted
Subgraph is extracted edge gradient rectangular histogram, and every 5 degree is an interval, the gradient in each interval range cumulative, by rectangular histogram normalization, forms the characteristic vector of 72 dimensions；
3. SIFT feature point rectangular histogram based on word bag model is extracted
Extract all subgraph SIFT feature vector；Use the K means clustering algorithm SIFT feature vector clusters to training set data, obtain 64 cluster centres, as the code book of word bag model；By all SIFT feature vector projections of each subgraph to code book, form the rectangular histogram of 64 dimensions and do normalization, obtaining characteristic vector；
4. three of the above characteristic vector is contacted, form the characteristic vector of 208 last dimensions.
(4) SVM classifier of off-line training mark, inputs the training of SVM instrument by the image feature vector of the positive negative sample of training set, and in training, positive and negative collection number of samples is all higher than 1000 herein, and SVM selects kernel function based on card side's distance；The training of this step obtains the grader for this TV programme special sign.
(5) subgraph to video to be split extracts the image feature vector identical with step (3), totally 208 dimension；Wherein, the code book of the needs forming the histogram feature vector of SIFT is the code book used in step (3), training set obtain through K Mean Method cluster.
(6) characteristic vector that step (5) is obtained by the SVM classifier using step (4) to obtain is classified, and classification results demarcates whether each subgraph exists mark.
(7) checked that if greater than equal to 3, then this camera lens of labelling is mark camera lens, and otherwise this camera lens of labelling is non-mark camera lens containing the number of key frames indicated in each camera lens by step (6) result.
(8) check the camera lens labelling of video to be split by camera lens, if adjacent two camera lens labellings are different, then using the border of the two camera lens as a cut-point, until the complete all adjacent camera lenses of sequential search, finally this program video inside division completes.
Claims (1)
1. utilizing the method that Mark Detection carries out TV programme automatic inside segmentation, its feature is, the method includes:
Step one, utilizes a kind of existing shot segmentation technique that television program video is carried out shot segmentation, it is thus achieved that the shot sequence information of Time Continuous；
Step 2, takes 5 frame key frames, and extracts the subgraph of the rectangular area of special sign position in all key frames each camera lens temporally average mode；
Step 3, extracts the image feature vector of all subgraphs of training set, and the subgraph containing mark is positive sample, and the subgraph without mark is negative sample, and training obtains SVM classifier；
Step 4, to this program video to be split, obtains all subgraphs through step one and step 2, extracts the image feature vector identical with step 3, and the SVM classifier obtained by step 3 is classified, and obtains the classification results of each subgraph；
Step 5, labelling camera lens flag property, if at least 3 frame subgraphs are judged as existing mark in camera lens, then this camera lens of labelling is mark camera lens, is otherwise labeled as non-mark camera lens；
Step 6, program video inside division, video has the adjacent shot boundary of unlike signal attribute as cut-point, Video segmentation is become paragraph；
Wherein, described step 2 specifically includes:
Step 1, is temporally divided into 6 sections by each camera lens, takes 5 two field pictures of adjacent segment as key frame；
Step 2, for these TV programme, determines the rectangular area at its place to known special sign, and mark is surrounded by this rectangle just completely, rectangular coordinates be (x, y, w, h), wherein x, y are respectively the transverse and longitudinal coordinate of rectangle upper left angle point, and w, h are respectively width and the height of rectangle；
Step 3, to this rectangle of all key-frame extraction, referred to as subgraph；
Wherein, described step 3 specifically includes:
Step 1, extracts hsv color statistic histogram to subgraph, and wherein H space is divided into 8 intervals, and S space is divided into 3 intervals, and V space is divided into 3 spaces, by rectangular histogram normalization, forms the characteristic vector of 72 dimensions；
Step 2, extracts edge gradient rectangular histogram to subgraph, and every 5 degree is an interval, the gradient in each interval range cumulative, by rectangular histogram normalization, forms the characteristic vector of 72 dimensions；
Step 3, extract all subgraph SIFT feature vector, use the K means clustering algorithm SIFT feature vector clusters to training set data, obtain 64 cluster centres, as code book, by all SIFT feature vector projections of each subgraph to code book, form the rectangular histogram of 64 dimensions and do normalization, obtaining characteristic vector；
Step 4, contacts three of the above characteristic vector, forms the characteristic vector of 208 last dimensions；
Step 5, uses the characteristic vector training SVM classifier of training set sample, and in training, positive and negative collection number of samples is all higher than 1000 herein, and SVM selects kernel function based on card side's distance, the training of this step to obtain the grader for this TV programme special sign.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010574074.8A CN102098449B (en) | 2010-12-06 | 2010-12-06 | A kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010574074.8A CN102098449B (en) | 2010-12-06 | 2010-12-06 | A kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102098449A CN102098449A (en) | 2011-06-15 |
| CN102098449B true CN102098449B (en) | 2016-08-03 |
Family
ID=44131296
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201010574074.8A Expired - Fee Related CN102098449B (en) | 2010-12-06 | 2010-12-06 | A kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102098449B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102436575A (en) * | 2011-09-22 | 2012-05-02 | Tcl集团股份有限公司 | Method for automatically detecting and classifying station captions |
| CN102799637A (en) * | 2012-06-27 | 2012-11-28 | 北京邮电大学 | Method for automatically generating main character abstract in television program |
| CN103034860A (en) * | 2012-12-14 | 2013-04-10 | 南京思创信息技术有限公司 | Scale-invariant feature transform (SIFT) based illegal building detection method |
| CN104185088B (en) * | 2014-03-03 | 2017-05-31 | 无锡天脉聚源传媒科技有限公司 | A kind of method for processing video frequency and device |
| CN105868768A (en) * | 2015-01-20 | 2016-08-17 | 阿里巴巴集团控股有限公司 | Method and system for recognizing whether picture carries specific marker |
| CN108270946A (en) * | 2016-12-30 | 2018-07-10 | 央视国际网络无锡有限公司 | A kind of computer-aided video editing device in feature based vector library |
| WO2018137126A1 (en) * | 2017-01-24 | 2018-08-02 | 深圳大学 | Method and device for generating static video abstract |
| CN109525901B (en) * | 2018-11-27 | 2020-08-25 | Oppo广东移动通信有限公司 | Video processing method and device, electronic equipment and computer readable medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5146336A (en) * | 1989-09-25 | 1992-09-08 | Le Groupe Videotron Ltee | Sync control for video overlay |
| CN101867729A (en) * | 2010-06-08 | 2010-10-20 | 上海交通大学 | Method for detecting news video formal soliloquy scene based on features of characters |
-
2010
- 2010-12-06 CN CN201010574074.8A patent/CN102098449B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5146336A (en) * | 1989-09-25 | 1992-09-08 | Le Groupe Videotron Ltee | Sync control for video overlay |
| CN101867729A (en) * | 2010-06-08 | 2010-10-20 | 上海交通大学 | Method for detecting news video formal soliloquy scene based on features of characters |
Non-Patent Citations (1)
| Title |
|---|
| 基于支持向量机和不变矩的交通标志检测;高联雄等;《计算机工程与应用》;20081101(第31期);第233-238页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102098449A (en) | 2011-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102098449B (en) | A kind of method utilizing Mark Detection to carry out TV programme automatic inside segmentation | |
| Wei et al. | Multi-vehicle detection algorithm through combining Harr and HOG features | |
| CN107346420B (en) | Character detection and positioning method in natural scene based on deep learning | |
| WO2017190656A1 (en) | Pedestrian re-recognition method and device | |
| CN112153483B (en) | Information implantation area detection method and device and electronic equipment | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| CN105631039A (en) | Picture browsing method | |
| Moradi et al. | Farsi/Arabic text extraction from video images by corner detection | |
| Zhang et al. | A multi-feature fusion based traffic light recognition algorithm for intelligent vehicles | |
| Xu et al. | Graphics and scene text classification in video | |
| Zhang et al. | Coarse-to-fine object detection in unmanned aerial vehicle imagery using lightweight convolutional neural network and deep motion saliency | |
| Bai et al. | Scene text localization using gradient local correlation | |
| Razalli et al. | Emergency vehicle recognition and classification method using HSV color segmentation | |
| Ye et al. | Jersey number detection in sports video for athlete identification | |
| CN114332956A (en) | Cattle face detection and cattle face key point positioning method based on convolutional neural network | |
| Song et al. | A novel image text extraction method based on k-means clustering | |
| Messelodi et al. | Scene text recognition and tracking to identify athletes in sport videos | |
| CN104463232A (en) | Density crowd counting method based on HOG characteristic and color histogram characteristic | |
| Zhu et al. | Detecting natural scenes text via auto image partition, two-stage grouping and two-layer classification | |
| CN104463134A (en) | License plate detection method and system | |
| CN107357834A (en) | Image retrieval method based on visual saliency fusion | |
| Li et al. | The research on traffic sign recognition based on deep learning | |
| Chen et al. | Effective candidate component extraction for text localization in born-digital images by combining text contours and stroke interior regions | |
| Shi et al. | Graph-based background suppression for scene text detection | |
| Chang | Intelligent text detection and extraction from natural scene images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160803 Termination date: 20211206 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |