CN106548118A - The recognition and retrieval method and system of cinema projection content - Google Patents
The recognition and retrieval method and system of cinema projection content Download PDFInfo
- Publication number
- CN106548118A CN106548118A CN201510613763.8A CN201510613763A CN106548118A CN 106548118 A CN106548118 A CN 106548118A CN 201510613763 A CN201510613763 A CN 201510613763A CN 106548118 A CN106548118 A CN 106548118A
- Authority
- CN
- China
- Prior art keywords
- retrieval
- similarity
- user
- matching
- query
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/732—Query formulation
- G06F16/7328—Query by example, e.g. a complete video frame or video sequence
-
- 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
- G06V20/48—Matching video sequences
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to the recognition and retrieval method and system of a kind of cinema projection content, are to solve the low problem of prior art identification recall precision, which in turn includes the following steps：1) media data step：Camera lens collect from fluorescent screen in video flowing in the sequence of video images that obtains；2) information retrieval step：User is extracted from video sequence interested, be adapted to the feature that retrieval is required；3) key frame retrieval step：Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted；4) images match step：The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity；5) browse retrieval step：It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query and key frame retrieval, retrieve the result for returning and may browse through.With being easily achieved, the advantage of cinema projection content recognition recall precision is significantly improved.
Description
Technical field
The present invention relates to a kind of cinema projection content identification method, the recognition and retrieval method and system of more particularly to a kind of cinema projection content.
Background technology
As the multi-medium datas such as developing rapidly for technology, image, video have been increasingly becoming information medium form main in field of information processing.Multi-medium data (Multimedia Data) refers to the carrier of various style informations, for example：The data such as text, figure, image, sound.It is characterized in：(1) multimedia data kind various (being unstructured data mostly), from different media, with diverse form and form；(2) multi-medium data amount is huge；(3) multi-medium data has time response and version concept, and the synchronized relation between media and inside media in time is such as must account in video on-demand system.The retrieval for being currently based on content mainly has three directions：Video (Video), audio frequency (Audio), image (Image).
So-called content-based retrieval (CBR, Content Based Retrieval), to refer to and enter line retrieval according to the semanteme and contextual relation of media object, it has following features：
■ extracts information clue from media content.Content-based retrieval breaches traditional limitation retrieved based on expression formula, and it is directly analyzed to image, video, audio frequency, extraction feature.Set up using these content characteristics and indexed into line retrieval.
■ content-based retrievals are a kind of approximate matches.During retrieval, it obtains the result of inquiry using the method Stepwise Refinement of similarity matching, i.e., inquiry is an iterative process, constantly reduces the scope of Query Result, until navigating to target.This point has significantly different with the fine matching method of routine data library searching.
The quick-searching of ■ large databases (collection).In actual multimedia database (collection), not only data volume is huge, and type and quantity are various, therefore it is required that CBR technologies also can rapidly realize the retrieval to multimedia messages as conventional information retrieval technique.
Content-based retrieval to be utilized the certain methods in the subjects such as image procossing, pattern recognition, computer vision, image understanding as part basis technology.CBR is not exclusively based on content, and be an information retrieval technique, it is inspired from the fields such as Cognitive Science, user model, image procossing, pattern recognition, knowledge base system, computer graphicss, data base management system and information retrieval, result in the expression and data model, effective and reliable Query Processing Algorithm, the generation of intelligent query interface and the retrieval technique unrelated with field and system structure of new media data.
One important research field of multimedia database content based video retrieval system is namely based on the video frequency searching of content.In multi-medium data, video data occupies very big proportion, and digitized video application in all fields is more and more universal, and has substantial amounts of video information to produce daily.People carry out system administration to which and aspect is quickly retrieved and proposes requirement.Video data, can be noted down due to it, retaining space and temporal various information so that people can obtain more detail contents more to approach natural mode.It information source is not carried out it is excessive abstract, although so as to feature things characteristic information in a certain respect unlike text description, lost many concrete details information.Therefore, application of the video data in the fields such as social life, medical treatment, military affairs is more and more extensive.
The content of the invention
Present invention aim to overcome that the drawbacks described above of prior art, there is provided one kind is easily achieved, the recognition and retrieval method of the cinema projection content of identification recall precision is significantly improved, and the object of the invention also resides in offer system for carrying out the process.
For achieving the above object, the recognition and retrieval method of cinema projection content of the present invention in turn includes the following steps：1) media data step：Camera lens collect from fluorescent screen in video flowing in the sequence of video images that obtains；2) information retrieval step：User is extracted from video sequence interested, be adapted to the feature that retrieval is required；3) key frame retrieval step：Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted；4) images match step：The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity；5) browse retrieval step：It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.With being easily achieved, the advantage of cinema projection content recognition recall precision is significantly improved.
Used as optimization, the query and search is content based video retrieval system and inquiry, is the process of a Stepwise Refinement, there is a Character adjustment, the cyclic process for matching again.
As optimization, the process：A initial query explanations：When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature；B, similarity matching：Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set；C, one group of candidate result for meeting certain similarity return to user by the big minispread of similarity；D, Character adjustment：One group returned to system meets the Query Result of initial characteristicses；E, query context is progressively reduced so, till user is satisfied with the result of inquiry.
Used as optimization, the video flowing of input is divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again for browsing and retrieval is used.
As optimization, in the case where memory capacity is limited, key frame is only stored；The method compared using rectangular histogram, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
As shown in figure 1, the function composition of its video frequency searching and explanation：Into the system operatio interface of Video content retrieval system, application program module and image retrieval matching module are entered via user's control module, on the one hand wherein application program module first carries out the acquisition of image sequence, after carry out the acquisition of video sequence, then carry out image sequence gray processing；On the other hand application program module first carries out the extraction of key frame, after carry out fitting a straight line and fitting of parabola, wherein image retrieval matching module carries out similarity algorithm operation.
As shown in Fig. 2 its workflow is：Collection video flowing, obtains media data, carries out information retrieval, then Jing images match and key frame retrieval, forms data and browse retrieval for user.Wherein：1. media data.The sequence of video images obtained from video flowing, all of analysis are all based on this data with processing.2. information retrieval.User is extracted from video sequence interested, be adapted to the feature that retrieval is required.3. key frame retrieval.Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted.4. images match.The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity.5. retrieval is browsed.It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.
As shown in figure 3, retrieving is user's query specification, similarity matching returns one group of candidate result, satisfied that an example is selected to end or from candidate result, changes query specification, re-starts similarity matching.Be content based video retrieval system and inquiry be a Stepwise Refinement process, there is a Character adjustment, the cyclic process for matching again：
1. initial query explanation.When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature.
2. similarity matching.Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set.
3. one group of candidate result for meeting certain similarity returns to user by the big minispread of similarity
4. Character adjustment.One group returned to system meets the Query Result of initial characteristicses.
5. query context is progressively reduced so, till user is satisfied with the result of inquiry.
The solution of technical problem：
1) separation of camera lens in order to be segmented to video sequence, must just be detected.It is simplest to be certainly identified with artificial mode, but efficiency is obviously very low.Automatically detected with computer, not only contribute to quickly split video, and also helped quickly classification.
2) it is to find the difference between lens image to the key of shot segmentation, the video flowing of input can be divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again can for browsing and retrieval is used.As the video data volume is huge, in the case where memory capacity is limited, key frame is generally only stored.The method for being compared using rectangular histogram herein, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
3) in view of Computer Vision speed and recall precision problem, therefore extract in HSI color model I luminance components make next step graphical analyses and process Data Source because I component is exactly 256 grades of greyscale colors of corresponding image.As HSI models mutually can be changed with RGB models, so important colouring information will not be lost, it may be necessary to be reduced to coloured image；And this is also only to have carried out the conversion of color in the experimental stage, the image procossing of true color is consistent with the process of gray level image in principle, so the retrieval of the key frame based on gray level image of this chapter discussion, images match and browsing and being equally applicable to coloured image.
The concept of grey level histogram and its application：The continuation method and discrete method of image procossing
There are two kinds of viewpoints when designing and realize the interpretative version of Digital Image Processing：People can regard digital picture as the set (practical situation is also such) of discrete sampling point, and each point is with its respective attribute.On the other hand, our images interested generally originate from physical world, and they obey the rule that can use continuous mathematics to describe well.In view of this consideration, image and its content Jing often can be preferably described with continuous function.
As digital picture is, based on discrete, therefore to adhere to continuous viewpoint simply and ignore discrete this fundamental characteristics to be dangerous.When the result analyzed when result and with continuous function has dramatically different, we term it sampling effect (Sampling Effect).As the equipment continuous function of the object in the scenery corresponding to image and imaging preferably can be represented, therefore thinking is confined to discrete mathematics and logical operationss is equally unadvisable.
In most of the cases, we are processing the image in the continuous world using discrete technology.The script state of image is continuous, and the result of process typically also will be deduced in a continuous fashion, image only when we using digital computer as instrument come to show our algorithm when, just temporarily become discrete form.So even if image is to be supplied to ours in digital form, we can not ignore its continuous basis.
The way taken is summarized as follows by we：First, it is therefore desirable to be able to portray the impact produced by carrying out after digitized to the image for being originally conitnuous forms.Second, we seek in by analog to digital transformation process again by digital to analogy, it is ensured that our contents interested are not lost or the not method of significantly sacrificing.3rd, it is therefore desirable to be able to prediction samples effect, can recognize that when they occur, and effective step can be taken to eliminate them or be reduced to the stage that can be tolerated.Continuous and discrete processing procedure is integrated into into a more typically method thus.
Peak detection based on Euclidean distance：
Pattern recognition and the purpose of pattern recognition：According to the definition of broad sense, pattern is some for imitation, perfect specimen.Pattern recognition, exactly identifies the specimen imitated by specific object.Identification ability is a kind of base attribute of the mankind and other biological, identification activity can be divided into two main Types according to the property of identified object：Specific object and abstract object.
According to the definition of narrow sense, pattern is the description of the quantitative or structure of object interested in some, and pattern class is the set of the pattern with some common denominators.Pattern recognition is a kind of automatic technique of research, and by this technology, machine automatically (or people as far as possible few interference) will be assigned to respective pattern apoplexy due to endogenous wind and go knowledge pattern is treated.
The purpose of research and development pattern recognition, is the perception for improving computer, so as to open up significantly computer application；And the real raising of computer perception, it is not only relevant with pattern recognition this subject itself, and the architecture with mathematics, engineering technology science and computer itself, soft hardware performance are relevant, we should draw the New TownMovement of fraternal subject in time, hold the new development of computer science, use by oneself, be that the mankind benefit.
Peak detection based on minimum distance classifier：Using the principle of the distance classifier of above-mentioned statistical-simulation spectrometry, using Euclidean distance minimum distance classifier method retrieving key frame.And in the Euclidean distance rectangular histogram of the difference of continuous frame sequence, key frame how being detected, the determination of this dynamic threshold is extremely important.
Such as Fig. 4, shown in the program flow diagram of crucial Frame Detection Algorithm, wherein the program circuit of crucial Frame Detection Algorithm is beginning, boundary value judgement, record current time, 1. determines the Euclidean distance value of continuous frame sequence, be stored in array FDistInt；2. maximum is chosen from every continuous three distance values of array FDistInt be stored in Palette arrays；3. Palette in array is sorted from small to large；4. the average of each element in Palette arrays is calculated, and using the average as dynamic threshold.5. the corresponding two field picture of element in all FDistInt arrays more than the threshold value is key frame.
For realizing that the system of the method for the invention includes user's control module, application program library module and image retrieval matching module；Wherein user's control module：It is responsible for providing the user the operating platform of control whole system running：Dispatch and using each function program in application program library module；Display to the user that computing and retrieval result；Application program library module：The Data Structures that responsible storage system needs, global variable and the application program towards specific tasks；Image retrieval matching module：Using similarity algorithm to the known image for providing, a most like two field picture is retrieved from key frame, and key frame is arranged from big to small according to similarity；
With following job step：1) media data step：Camera lens collect from fluorescent screen in video flowing in the sequence of video images that obtains；2) information retrieval step：User is extracted from video sequence interested, be adapted to the feature that retrieval is required；3) key frame retrieval step：Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted；4) images match step：The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity；5) browse retrieval step：It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.With being easily achieved, the advantage of cinema projection content recognition recall precision is significantly improved.
Used as optimization, the query and search is content based video retrieval system and inquiry, is the process of a Stepwise Refinement, there is a Character adjustment, the cyclic process for matching again.
As optimization, the process：A initial query explanations：When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature；B, similarity matching：Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set；C, one group of candidate result for meeting certain similarity return to user by the big minispread of similarity；D, Character adjustment：One group returned to system meets the Query Result of initial characteristicses；E, query context is progressively reduced so, till user is satisfied with the result of inquiry.
Used as optimization, the video flowing of input is divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again for browsing and retrieval is used.
As optimization, in the case where memory capacity is limited, key frame is only stored；The method compared using rectangular histogram, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
As shown in figure 1, the function composition of its video frequency searching and explanation：Into the system operatio interface of Video content retrieval system, application program module and image retrieval matching module are entered via user's control module, on the one hand wherein application program module first carries out the acquisition of image sequence, after carry out the acquisition of video sequence, then carry out image sequence gray processing；On the other hand application program module first carries out the extraction of key frame, after carry out fitting a straight line and fitting of parabola, wherein image retrieval matching module carries out similarity algorithm operation.
As shown in Fig. 2 its workflow is：Collection video flowing, obtains media data, carries out information retrieval, then Jing images match and key frame retrieval, forms data and browse retrieval for user.Wherein：1. media data.The sequence of video images obtained from video flowing, all of analysis are all based on this data with processing.2. information retrieval.User is extracted from video sequence interested, be adapted to the feature that retrieval is required.3. key frame retrieval.Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted.4. images match.The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity.5. retrieval is browsed.It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.
As shown in figure 3, retrieving is user's query specification, similarity matching returns one group of candidate result, satisfied that an example is selected to end or from candidate result, changes query specification, re-starts similarity matching.Be content based video retrieval system and inquiry be a Stepwise Refinement process, there is a Character adjustment, the cyclic process for matching again：
6. initial query explanation.When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature.
7. similarity matching.Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set.
8. one group of candidate result for meeting certain similarity returns to user by the big minispread of similarity
9. Character adjustment.One group returned to system meets the Query Result of initial characteristicses.
10. query context is progressively reduced so, till user is satisfied with the result of inquiry.
The solution of technical problem：
1) separation of camera lens in order to be segmented to video sequence, must just be detected.It is simplest to be certainly identified with artificial mode, but efficiency is obviously very low.Automatically detected with computer, not only contribute to quickly split video, and also helped quickly classification.
2) it is to find the difference between lens image to the key of shot segmentation, the video flowing of input can be divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again can for browsing and retrieval is used.As the video data volume is huge, in the case where memory capacity is limited, key frame is generally only stored.The method for being compared using rectangular histogram herein, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
3) in view of Computer Vision speed and recall precision problem, therefore extract in HSI color model I luminance components make next step graphical analyses and process Data Source because I component is exactly 256 grades of greyscale colors of corresponding image.As HSI models mutually can be changed with RGB models, so important colouring information will not be lost, it may be necessary to be reduced to coloured image；And this is also only to have carried out the conversion of color in the experimental stage, the image procossing of true color is consistent with the process of gray level image in principle, so the retrieval of the key frame based on gray level image of this chapter discussion, images match and browsing and being equally applicable to coloured image.
The concept of grey level histogram and its application：The continuation method and discrete method of image procossing
There are two kinds of viewpoints when designing and realize the interpretative version of Digital Image Processing：People can regard digital picture as the set (practical situation is also such) of discrete sampling point, and each point is with its respective attribute.On the other hand, our images interested generally originate from physical world, and they obey the rule that can use continuous mathematics to describe well.In view of this consideration, image and its content Jing often can be preferably described with continuous function.
As digital picture is, based on discrete, therefore to adhere to continuous viewpoint simply and ignore discrete this fundamental characteristics to be dangerous.When the result analyzed when result and with continuous function has dramatically different, we term it sampling effect (Sampling Effect).As the equipment continuous function of the object in the scenery corresponding to image and imaging preferably can be represented, therefore thinking is confined to discrete mathematics and logical operationss is equally unadvisable.
In most of the cases, we are processing the image in the continuous world using discrete technology.The script state of image is continuous, and the result of process typically also will be deduced in a continuous fashion, image only when we using digital computer as instrument to realize our algorithm when, just temporarily become discrete form.So even if image is to be supplied to ours in digital form, we can not ignore its continuous basis.
The way taken is summarized as follows by we：First, it is therefore desirable to be able to portray the impact produced by carrying out after digitized to the image for being originally conitnuous forms.Second, we seek in by analog to digital transformation process again by digital to analogy, it is ensured that our contents interested are not lost or the not method of significantly sacrificing.3rd, it is therefore desirable to be able to prediction samples effect, can recognize that when they occur, and effective step can be taken to eliminate them or be reduced to the stage that can be tolerated.Continuous and discrete processing procedure is integrated into into a more typically method thus.
Peak detection based on Euclidean distance：
Pattern recognition and the purpose of pattern recognition：According to the definition of broad sense, pattern is some for imitation, perfect specimen.Pattern recognition, exactly identifies the specimen imitated by specific object.Identification ability is a kind of base attribute of the mankind and other biological, identification activity can be divided into two main Types according to the property of identified object：Specific object and abstract object.
According to the definition of narrow sense, pattern is the description of the quantitative or structure of object interested in some, and pattern class is the set of the pattern with some common denominators.Pattern recognition is a kind of automatic technique of research, and by this technology, machine automatically (or people as far as possible few interference) will be assigned to respective pattern apoplexy due to endogenous wind and go knowledge pattern is treated.
The purpose of research and development pattern recognition, is the perception for improving computer, so as to open up significantly computer application；And the real raising of computer perception, it is not only relevant with pattern recognition this subject itself, and the architecture with mathematics, engineering technology science and computer itself, soft hardware performance are relevant, we should draw the New TownMovement of fraternal subject in time, hold the new development of computer science, use by oneself, be that the mankind benefit.
Peak detection based on minimum distance classifier：Using the principle of the distance classifier of above-mentioned statistical-simulation spectrometry, using Euclidean distance minimum distance classifier method retrieving key frame.And in the Euclidean distance rectangular histogram of the difference of continuous frame sequence, key frame how being detected, the determination of this dynamic threshold is extremely important.
Such as Fig. 4, shown in the program flow diagram of crucial Frame Detection Algorithm, wherein the program circuit of crucial Frame Detection Algorithm is beginning, boundary value judgement, record current time, 1. determines the Euclidean distance value of continuous frame sequence, be stored in array FDi stInt；2. maximum is chosen from every continuous three distance values of array FDistInt be stored in Palette arrays；3. Palette in array is sorted from small to large；4. the average of each element in Palette arrays is calculated, and using the average as dynamic threshold.5. the corresponding two field picture of element in all FDistInt arrays more than the threshold value is key frame.
After using above-mentioned technical proposal, the recognition and retrieval method and system of cinema projection content of the present invention significantly improve the advantage of cinema projection content recognition recall precision with being easily achieved.
Description of the drawings
Fig. 1 is the function composition of the video frequency searching of the recognition and retrieval method and system of cinema projection content of the present invention and illustrates block diagram；
Fig. 2 is the system architecture diagram of the audio content retrieval of the recognition and retrieval method and system of cinema projection content of the present invention；
Fig. 3 is the retrieving block diagram of the recognition and retrieval method and system of cinema projection content of the present invention；
Fig. 4 is the program flow diagram of the crucial Frame Detection Algorithm of the recognition and retrieval method and system of cinema projection content of the present invention.
Specific embodiment
The recognition and retrieval method of cinema projection content of the present invention in turn includes the following steps：1) media data step：Camera lens collect from fluorescent screen in video flowing in the sequence of video images that obtains；2) information retrieval step：User is extracted from video sequence interested, be adapted to the feature that retrieval is required；3) key frame retrieval step：Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted：4) images match step：The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity；5) browse retrieval step：It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.
Specifically described query and search is content based video retrieval system and inquiry, is the process of a Stepwise Refinement there is a Character adjustment, the cyclic process for matching again.
More specifically described process：A initial query explanations：When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature；B, similarity matching：Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set；C, one group of candidate result for meeting certain similarity return to user by the big minispread of similarity；D, Character adjustment：One group returned to system meets the Query Result of initial characteristicses；E, query context is progressively reduced so, till user is satisfied with the result of inquiry.
It is preferred that the video flowing of input to be divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again for browsing and retrieval is used.
More preferably in the case where memory capacity is limited, key frame is only stored；The method compared using rectangular histogram, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
As shown in figure 1, the function composition of its video frequency searching and explanation：Into the system operatio interface of Video content retrieval system, application program module and image retrieval matching module are entered via user's control module, on the one hand wherein application program module first carries out the acquisition of image sequence, after carry out the acquisition of video sequence, then carry out image sequence gray processing；On the other hand application program module first carries out the extraction of key frame, after carry out fitting a straight line and fitting of parabola, wherein image retrieval matching module carries out similarity algorithm operation.
As shown in Fig. 2 its workflow is：Collection video flowing, obtains media data, carries out information retrieval, then Jing images match and key frame retrieval, forms data and browse retrieval for user.Wherein：1. media data.The sequence of video images obtained from video flowing, all of analysis are all based on this data with processing.2. information retrieval.User is extracted from video sequence interested, be adapted to the feature that retrieval is required.3. key frame retrieval.Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted.4. images match.The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity.5. retrieval is browsed.It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.
As shown in figure 3, retrieving is user's query specification, similarity matching returns one group of candidate result, satisfied that an example is selected to end or from candidate result, changes query specification, re-starts similarity matching.Be content based video retrieval system and inquiry be a Stepwise Refinement process, there is a Character adjustment, the cyclic process for matching again：
11. initial query explanations.When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature.
12. similaritys are matched.Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set.
13. one group of candidate result for meeting certain similarity return to user by the big minispread of similarity
14. Character adjustments.One group returned to system meets the Query Result of initial characteristicses.
15. so progressively reduce query context, till user is satisfied with the result of inquiry.
The solution of technical problem：
1) separation of camera lens in order to be segmented to video sequence, must just be detected.It is simplest to be certainly identified with artificial mode, but efficiency is obviously very low.Automatically detected with computer, not only contribute to quickly split video, and also helped quickly classification.
2) it is to find the difference between lens image to the key of shot segmentation, the video flowing of input can be divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again can for browsing and retrieval is used.As the video data volume is huge, in the case where memory capacity is limited, key frame is generally only stored.The method for being compared using rectangular histogram herein, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
3) in view of Computer Vision speed and recall precision problem, therefore extract in HSI color model I luminance components make next step graphical analyses and process Data Source because I component is exactly 256 grades of greyscale colors of corresponding image.As HSI models mutually can be changed with RGB models, so important colouring information will not be lost, it may be necessary to be reduced to coloured image；And this is also only to have carried out the conversion of color in the experimental stage, the image procossing of true color is consistent with the process of gray level image in principle, so the retrieval of the key frame based on gray level image of this chapter discussion, images match and browsing and being equally applicable to coloured image.
The concept of grey level histogram and its application：The continuation method and discrete method of image procossing
There are two kinds of viewpoints when designing and realize the interpretative version of Digital Image Processing：People can regard digital picture as the set (practical situation is also such) of discrete sampling point, and each point is with its respective attribute.On the other hand, our images interested generally originate from physical world, and they obey the rule that can use continuous mathematics to describe well.In view of this consideration, image and its content Jing often can be preferably described with continuous function.
As digital picture is, based on discrete, therefore to adhere to continuous viewpoint simply and ignore discrete this fundamental characteristics to be dangerous.When the result analyzed when result and with continuous function has dramatically different, we term it sampling effect (Sampling Effect).As the equipment continuous function of the object in the scenery corresponding to image and imaging preferably can be represented, therefore thinking is confined to discrete mathematics and logical operationss is equally unadvisable.
In most of the cases, we are processing the image in the continuous world using discrete technology.The script state of image is continuous, and the result of process typically also will be deduced in a continuous fashion, image only when we using digital computer as instrument to realize our algorithm when, just temporarily become discrete form.So even if image is to be supplied to ours in digital form, we can not ignore its continuous basis.
The way taken is summarized as follows by we：First, it is therefore desirable to be able to portray the impact produced by carrying out after digitized to the image for being originally conitnuous forms.Second, we seek in by analog to digital transformation process again by digital to analogy, it is ensured that our contents interested are not lost or the not method of significantly sacrificing.3rd, it is therefore desirable to be able to prediction samples effect, can recognize that when they occur, and effective step can be taken to eliminate them or be reduced to the stage that can be tolerated.Continuous and discrete processing procedure is integrated into into a more typically method thus.
Peak detection based on Euclidean distance：
Pattern recognition and the purpose of pattern recognition：According to the definition of broad sense, pattern is some for imitation, perfect specimen.Pattern recognition, exactly identifies the specimen imitated by specific object.Identification ability is a kind of base attribute of the mankind and other biological, identification activity can be divided into two main Types according to the property of identified object：Specific object and abstract object.
According to the definition of narrow sense, pattern is the description of the quantitative or structure of object interested in some, and pattern class is the set of the pattern with some common denominators.Pattern recognition is a kind of automatic technique of research, and by this technology, machine automatically (or people as far as possible few interference) will be assigned to respective pattern apoplexy due to endogenous wind and go knowledge pattern is treated.
The purpose of research and development pattern recognition, is the perception for improving computer, so as to open up significantly computer application：And the real raising of computer perception, it is not only relevant with pattern recognition this subject itself, and the architecture with mathematics, engineering technology science and computer itself, soft hardware performance are relevant, we should draw the New TownMovement of fraternal subject in time, hold the new development of computer science, use by oneself, be that the mankind benefit.
Peak detection based on minimum distance classifier：Using the principle of the distance classifier of above-mentioned statistical-simulation spectrometry, using Euclidean distance minimum distance classifier method retrieving key frame.And in the Euclidean distance rectangular histogram of the difference of continuous frame sequence, key frame how being detected, the determination of this dynamic threshold is extremely important.
Such as Fig. 4, shown in the program flow diagram of crucial Frame Detection Algorithm, wherein the program circuit of crucial Frame Detection Algorithm is beginning, boundary value judgement, record current time, 1. determines the Euclidean distance value of continuous frame sequence, be stored in array FDistInt；2. maximum is chosen from every continuous three distance values of array FDistInt be stored in Palette arrays；3. Palette in array is sorted from small to large；4. the average of each element in Palette arrays is calculated, and using the average as dynamic threshold.5. the corresponding two field picture of element in all FDistInt arrays more than the threshold value is key frame.
After using above-mentioned technical proposal, the recognition and retrieval method and system of cinema projection content of the present invention significantly improve the advantage of cinema projection content recognition recall precision with being easily achieved.
For realizing that the system of the method for the invention includes user's control module, application program library module and image retrieval matching module；Wherein user's control module：It is responsible for providing the user the operating platform of control whole system running；Dispatch and using each function program in application program library module；Display to the user that computing and retrieval result；Application program library module：The Data Structures that responsible storage system needs, global variable and the application program towards specific tasks；Image retrieval matching module：Using similarity algorithm to the known image for providing, a most like two field picture is retrieved from key frame, and key frame is arranged from big to small according to similarity.
With following job step：1) media data step：Camera lens collect from fluorescent screen in video flowing in the sequence of video images that obtains；2) information retrieval step：User is extracted from video sequence interested, be adapted to the feature that retrieval is required；3) key frame retrieval step：Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted；4) images match step：The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity；5) browse retrieval step：It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.
Specifically described query and search is content based video retrieval system and inquiry, is the process of a Stepwise Refinement there is a Character adjustment, the cyclic process for matching again.
More specifically described process：A initial query explanations：When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature；B, similarity matching：Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set；C, one group of candidate result for meeting certain similarity return to user by the big minispread of similarity；D, Character adjustment：One group returned to system meets the Query Result of initial characteristicses；E, query context is progressively reduced so, till user is satisfied with the result of inquiry.
It is preferred that the video flowing of input to be divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again for browsing and retrieval is used.
More preferably in the case where memory capacity is limited, key frame is only stored：The method compared using rectangular histogram, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
As shown in figure 1, the function composition of its video frequency searching and explanation：Into the system operatio interface of Video content retrieval system, application program module and image retrieval matching module are entered via user's control module, on the one hand wherein application program module first carries out the acquisition of image sequence, after carry out the acquisition of video sequence, then carry out image sequence gray processing；On the other hand application program module first carries out the extraction of key frame, after carry out fitting a straight line and fitting of parabola, wherein image retrieval matching module carries out similarity algorithm operation.
As shown in Fig. 2 its workflow is：Collection video flowing, obtains media data, carries out information retrieval, then Jing images match and key frame retrieval, forms data and browse retrieval for user.Wherein：1. media data.The sequence of video images obtained from video flowing, all of analysis are all based on this data with processing.2. information retrieval.User is extracted from video sequence interested, be adapted to the feature that retrieval is required.3. key frame retrieval.Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted.4. images match.The matching of the example image to giving carries out similarity retrieval using distance function, and matching result is arranged from big to small according to similarity.5. retrieval is browsed.It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query (QBE) and key frame retrieval, retrieve the result for returning and may browse through.
As shown in figure 3, retrieving is user's query specification, similarity matching returns one group of candidate result, satisfied that an example is selected to end or from candidate result, changes query specification, re-starts similarity matching.Be content based video retrieval system and inquiry be a Stepwise Refinement process, there is a Character adjustment, the cyclic process for matching again：
16. initial query explanations.When user searches an object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific feature.
17. similaritys are matched.Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set.
18. one group of candidate result for meeting certain similarity return to user by the big minispread of similarity
19. Character adjustments.One group returned to system meets the Query Result of initial characteristicses.
20. so progressively reduce query context, till user is satisfied with the result of inquiry.
The solution of technical problem：
1) separation of camera lens in order to be segmented to video sequence, must just be detected.It is simplest to be certainly identified with artificial mode, but efficiency is obviously very low.Automatically detected with computer, not only contribute to quickly split video, and also helped quickly classification.
2) it is to find the difference between lens image to the key of shot segmentation, the video flowing of input can be divided into shot boundary detector the set of its elementary cell-camera lens；Key frame and movable information are extracted on this basis again can for browsing and retrieval is used.As the video data volume is huge, in the case where memory capacity is limited, key frame is generally only stored.The method for being compared using rectangular histogram herein, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, it is proposed that the crucial Frame Detection Algorithm based on histogram analysis of optimization.
3) in view of Computer Vision speed and recall precision problem, therefore extract in HSI color model I luminance components make next step graphical analyses and process Data Source because I component is exactly 256 grades of greyscale colors of corresponding image.As HSI models mutually can be changed with RGB models, so important colouring information will not be lost, it may be necessary to be reduced to coloured image；And this is also only to have carried out the conversion of color in the experimental stage, the image procossing of true color is consistent with the process of gray level image in principle, so the retrieval of the key frame based on gray level image of this chapter discussion, images match and browsing and being equally applicable to coloured image.
The concept of grey level histogram and its application：The continuation method and discrete method of image procossing
There are two kinds of viewpoints when designing and realize the interpretative version of Digital Image Processing：People can regard digital picture as the set (practical situation is also such) of discrete sampling point, and each point is with its respective attribute.On the other hand, our images interested generally originate from physical world, and they obey the rule that can use continuous mathematics to describe well.In view of this consideration, image and its content Jing often can be preferably described with continuous function.
As digital picture is, based on discrete, therefore to adhere to continuous viewpoint simply and ignore discrete this fundamental characteristics to be dangerous.When result and with continuous function analyze result row it is dramatically different when, we term it sampling effect (Sampling Effect).As the equipment continuous function of the object in the scenery corresponding to image and imaging preferably can be represented, therefore thinking is confined to discrete mathematics and logical operationss is equally unadvisable.
In most of the cases, we are processing the image in the continuous world using discrete technology.The script state of image is continuous, and the result of process typically also will be deduced in a continuous fashion, image only when we using digital computer as instrument to realize our algorithm when, just temporarily become discrete form.So even if image is to be supplied to ours in digital form, we can not ignore its continuous basis.
The way taken is summarized as follows by we：First, it is therefore desirable to be able to portray the impact produced by carrying out after digitized to the image for being originally conitnuous forms.Second, we seek in by analog to digital transformation process again by digital to analogy, it is ensured that our contents interested are not lost or the not method of significantly sacrificing.3rd, it is therefore desirable to be able to prediction samples effect, can recognize that when they occur, and effective step can be taken to eliminate them or be reduced to the stage that can be tolerated.Continuous and discrete processing procedure is integrated into into a more typically method thus.
Peak detection based on Euclidean distance：
Pattern recognition and the purpose of pattern recognition：According to the definition of broad sense, pattern is some for imitation, perfect specimen.Pattern recognition, exactly identifies the specimen imitated by specific object.Identification ability is a kind of base attribute of the mankind and other biological, identification activity can be divided into two main Types according to the property of identified object：Specific object and abstract object.
According to the definition of narrow sense, pattern is the description of the quantitative or structure of object interested in some, and pattern class is the set of the pattern with some common denominators.Pattern recognition is a kind of automatic technique of research, and by this technology, machine automatically (or people as far as possible few interference) will be assigned to respective pattern apoplexy due to endogenous wind and go knowledge pattern is treated.
The purpose of research and development pattern recognition, is the perception for improving computer, so as to open up significantly computer application；And the real raising of computer perception, it is not only relevant with pattern recognition this subject itself, and the architecture with mathematics, engineering technology science and computer itself, soft hardware performance are relevant, we should draw the New TownMovement of fraternal subject in time, hold the new development of computer science, use by oneself, be that the mankind benefit.
Peak detection based on minimum distance classifier：Using the principle of the distance classifier of above-mentioned statistical-simulation spectrometry, using Euclidean distance minimum distance classifier method retrieving key frame.And in the Euclidean distance rectangular histogram of the difference of continuous frame sequence, key frame how being detected, the determination of this dynamic threshold is extremely important.
Such as Fig. 4, shown in the program flow diagram of crucial Frame Detection Algorithm, wherein the program circuit of crucial Frame Detection Algorithm is beginning, boundary value judgement, record current time, 1. determines the Euclidean distance value of continuous frame sequence, be stored in array FDistInt；2. maximum is chosen from every continuous three distance values of array FDistInt be stored in Palette arrays；3. Palette in array is sorted from small to large；4. the average of each element in Palette arrays is calculated, and using the average as dynamic threshold.5. the corresponding two field picture of element in all FDistInt arrays more than the threshold value is key frame.
After using above-mentioned technical proposal, the recognition and retrieval method and system of cinema projection content of the present invention significantly improve the advantage of cinema projection content recognition recall precision with being easily achieved.
Claims (10)
1. a kind of recognition and retrieval method of cinema projection content, it is characterised in that in turn include the following steps：1) media data
Step：Camera lens collect from fluorescent screen in video flowing in the sequence of video images that obtains；2) information retrieval step：From video sequence
Extraction user is interested, is adapted to the feature that retrieval is required；3) key frame retrieval step：Using the searching algorithm of various key frames
Shot segmentation is carried out to video sequence, extract keyframe sequence；4) images match step：To give example image
Similarity retrieval is carried out with using distance function, and matching result is arranged from big to small according to similarity；5) browse retrieval step：
It is main that Retrieval Interface is provided a user with the Visual Inquiry such as sample query and key frame retrieval, retrieve the result for returning and may browse through.
2. method according to claim 1, it is characterised in that the query and search be content based video retrieval system and inquiry,
It is the process of a Stepwise Refinement, there is a Character adjustment, the cyclic process for matching again.
3. method according to claim 2, it is characterised in that the process：A initial query explanations：User searches one
During object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific spy
Levy；B, similarity matching：Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set；
C, one group of candidate result for meeting certain similarity return to user by the big minispread of similarity；D, Character adjustment：To system
One group for returning meets the Query Result of initial characteristicses；E, query context is progressively reduced so, until user is satisfied with the knot of platform inquiry
Till fruit.
4. according to the arbitrary methods described of claim 1-3, it is characterised in that the video flowing of input is divided into by shot boundary detector
The set of its one camera lens of elementary cell；Key frame and movable information are extracted on this basis again for browsing and retrieval is used.
5. method according to claim 4, it is characterised in that in the case where memory capacity is limited, only store key frame；
The method compared using rectangular histogram, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, is proposed
The crucial Frame Detection Algorithm based on histogram analysis of optimization.
6. the system for realizing claim 1 methods described is used for, it is characterised in that including user's control module, application library
Module and image retrieval matching module；Wherein user's control module：It is responsible for providing the user the operation of control whole system running
Platform；Dispatch and using each function program in application program library module；Display to the user that computing and retrieval result；Using
Library module：The Data Structures that responsible storage system needs, global variable and the application program towards specific tasks；
Image retrieval matching module：Using similarity algorithm to the known image for providing, a most like frame is retrieved from key frame
Image, and key frame is arranged from big to small according to similarity；
With following job step：1) media data step：Camera lens collect from fluorescent screen in video flowing in the video image that obtains
Sequence；2) information retrieval step：User is extracted from video sequence interested, be adapted to the feature that retrieval is required；3) it is crucial
Frame retrieval step：Shot segmentation is carried out to video sequence using the searching algorithm of various key frames, keyframe sequence is extracted；4)
Images match is walked：The matching of the example image to giving carries out similarity retrieval using distance function, and by matching result according to
Similarity is arranged from big to small；5) browse retrieval step：It is main to be carried to user with the Visual Inquiry such as sample query and key frame retrieval
For Retrieval Interface, retrieve the result for returning and may browse through.
7. system according to claim 6, it is characterised in that the query and search be content based video retrieval system and inquiry,
It is the process of a Stepwise Refinement, there is a Character adjustment, the cyclic process for matching again.
8. system according to claim 7, it is characterised in that the process：A initial query explanations：User searches one
During object, can initially provide feature to form an inquiry, system is extracted feature or query specification is mapped as specific spy
Levy；B, similarity matching：Query characteristics are carried out into Similarity matching according to certain matching algorithm with the feature in key frame set；
C, one group of candidate result for meeting certain similarity return to user by the big minispread of similarity；D, Character adjustment：To system
One group for returning meets the Query Result of initial characteristicses：E, query context is progressively reduced so, until user is satisfied with the knot of inquiry
Till fruit.
9. according to the arbitrary system of claim 6-8, it is characterised in that the video flowing of input is divided into by shot boundary detector
The set of its one camera lens of elementary cell；Key frame and movable information are extracted on this basis again for browsing and retrieval is used.
10. system according to claim 9, it is characterised in that in the case where memory capacity is limited, only store key frame；
The method compared using rectangular histogram, on the basis of the Algorithm Analysis compared to traditional rectangular histogram, comparing, checking, is proposed
The crucial Frame Detection Algorithm based on histogram analysis of optimization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510613763.8A CN106548118A (en) | 2015-09-23 | 2015-09-23 | The recognition and retrieval method and system of cinema projection content |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510613763.8A CN106548118A (en) | 2015-09-23 | 2015-09-23 | The recognition and retrieval method and system of cinema projection content |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106548118A true CN106548118A (en) | 2017-03-29 |
Family
ID=58365236
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510613763.8A Pending CN106548118A (en) | 2015-09-23 | 2015-09-23 | The recognition and retrieval method and system of cinema projection content |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106548118A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109189991A (en) * | 2018-08-17 | 2019-01-11 | 百度在线网络技术（北京）有限公司 | Repeat video frequency identifying method, device, terminal and computer readable storage medium |
| CN113312521A (en) * | 2021-06-15 | 2021-08-27 | 北京百度网讯科技有限公司 | Content retrieval method, device, electronic equipment and medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101064846A (en) * | 2007-05-24 | 2007-10-31 | 上海交通大学 | Time-shifted television video matching method combining program content metadata and content analysis |
-
2015
- 2015-09-23 CN CN201510613763.8A patent/CN106548118A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101064846A (en) * | 2007-05-24 | 2007-10-31 | 上海交通大学 | Time-shifted television video matching method combining program content metadata and content analysis |
Non-Patent Citations (1)
| Title |
|---|
| 张若英: "基于内容的视频检索方法的研究及实验系统的开发", 《豆丁网在线公开：HTTPS://WWW.DOCIN.COM/P-727108923.HTML》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109189991A (en) * | 2018-08-17 | 2019-01-11 | 百度在线网络技术（北京）有限公司 | Repeat video frequency identifying method, device, terminal and computer readable storage medium |
| CN109189991B (en) * | 2018-08-17 | 2021-06-08 | 百度在线网络技术（北京）有限公司 | Duplicate video identification method, device, terminal and computer readable storage medium |
| CN113312521A (en) * | 2021-06-15 | 2021-08-27 | 北京百度网讯科技有限公司 | Content retrieval method, device, electronic equipment and medium |
| CN113312521B (en) * | 2021-06-15 | 2023-08-01 | 北京百度网讯科技有限公司 | Content retrieval method, device, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cheng et al. | Salientshape: group saliency in image collections | |
| JP4981221B2 (en) | Media segmentation system and related methods | |
| Fei et al. | Memorable and rich video summarization | |
| CN110442747B (en) | Video abstract generation method based on keywords | |
| Priya et al. | Shot based keyframe extraction for ecological video indexing and retrieval | |
| Meng et al. | Object instance search in videos via spatio-temporal trajectory discovery | |
| dos Santos Belo et al. | Summarizing video sequence using a graph-based hierarchical approach | |
| CN110751232A (en) | Chinese complex scene text detection and identification method | |
| Atlam et al. | Comparative study on CBIR based on color feature | |
| Vimina et al. | A sub-block based image retrieval using modified integrated region matching | |
| Fei et al. | Creating memorable video summaries that satisfy the user’s intention for taking the videos | |
| CN110287369B (en) | Semantic-based video retrieval method and system | |
| Zhu et al. | Video scene segmentation and semantic representation using a novel scheme | |
| CN110765314A (en) | Video semantic structural extraction and labeling method | |
| CN106548118A (en) | The recognition and retrieval method and system of cinema projection content | |
| JP2009060413A (en) | Method and system for extracting feature of moving image, and method and system for retrieving moving image | |
| Brindha et al. | Bridging semantic gap between high-level and low-level features in content-based video retrieval using multi-stage ESN–SVM classifier | |
| Memon et al. | Region based localized matching image retrieval system using color-size features for image retrieval | |
| Rudinac et al. | Comparison of cbir systems with different number of feature vector components | |
| Anh et al. | Video retrieval using histogram and sift combined with graph-based image segmentation | |
| Chatur et al. | A simple review on content based video images retrieval | |
| Paul et al. | Issues in database management of multimedia information | |
| Heesch et al. | Video Retrieval Using Search and Browsing with Key Frames. | |
| Shambharkar et al. | A comparative study on retrieved images by content based image retrieval system based on binary tree, color, texture and canny edge detection approach | |
| Gudewar et al. | Ontology to improve CBIR system |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170329 |