CN104778224B - A kind of destination object social networks recognition methods based on video semanteme - Google Patents

A kind of destination object social networks recognition methods based on video semanteme Download PDF

Info

Publication number
CN104778224B
CN104778224B CN201510137760.1A CN201510137760A CN104778224B CN 104778224 B CN104778224 B CN 104778224B CN 201510137760 A CN201510137760 A CN 201510137760A CN 104778224 B CN104778224 B CN 104778224B
Authority
CN
China
Prior art keywords
personage
node
camera lens
semantic
scene
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
Application number
CN201510137760.1A
Other languages
Chinese (zh)
Other versions
CN104778224A (en
Inventor
陈志�
高翔
岳文静
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.)
Fan Liyang
Li Bo
Original Assignee
Nanjing Post and Telecommunication University
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 Nanjing Post and Telecommunication University filed Critical Nanjing Post and Telecommunication University
Priority to CN201510137760.1A priority Critical patent/CN104778224B/en
Publication of CN104778224A publication Critical patent/CN104778224A/en
Application granted granted Critical
Publication of CN104778224B publication Critical patent/CN104778224B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

A kind of destination object social networks recognition methods based on video semanteme, the video data of user's input is pre-processed first, obtain lens image frame sequence, it is taken out key frame, the characteristic vector of the key frame is extracted by SVM learning models, the semantic storage camera lens node into XML file of camera lens that analysis characteristic vector is obtained, then the semantic node according to corresponding to the node lower time of each camera lens and personage, the camera lens node for possessing identical personage's semanteme node is classified as one group of camera lens node, the data for the every group of camera lens node classified are saved under the scene node of XML file according to the incremental order of the nodal value of entitled time node, camera lens semantic sequence is constructed successively, represent scene one by one;Finally with Scene Semantics matrix one by one come the semantic information and social networks of personage in storage scenarios, then by way of taking union, personage's semantic information in all Scene Semantics matrixes and social networks are merged into the matrix of a big representative video semanteme.

Description

A kind of destination object social networks recognition methods based on video semanteme
Technical field
The present invention relates to a kind of social networks recognition methods, by carrying out semantic analysis to video, what identification was wherein implied Social networks, belong to image procossing, social networks, the interleaving techniques field of software engineering.
Background technology
Social Media refers to website and the technology for allowing people write, share, evaluating, discussing, communicating with each other.It is people Be used for sharing the instrument and platform of opinion, opinion, experience and viewpoint each other, mainly include at this stage social network sites, microblogging, Wechat, blog etc..Social Media is a kind of cloud service, and cloud computing technology extensive use is inherently a kind of with Social Media Web applications based on cloud computing.
Social networks is social networking service, as image of one people of evolution silently of social activity on network more tends to Completely, at this time social networks occurs.External main representative product has Facebook, Twitter, and domestic main representative has Renren Network, happy net, Sina weibo etc..Instantly the hot spot technology of social networks is to combine cloud computing, ecommerce and emotion sense Know technology.Social networks refer to the interpersonal relationships in social communication, also refer to the relation between good friend in social networks.By user it Between contact and mutual dynamic frequency, social networks be divided to strong relation and the major class of weak relation two.
The structure of video can be generally divided into four levels from high toward low:Video sequence, scene, camera lens, frame.One video Sequence generally refers to a single video file, or a video segment.Video sequence is made up of several scenes.Each Scene includes one or more semantic related camera lenses, and these camera lenses can be continuous or spaced.Each camera lens bag Containing some continuous picture frames.Video semanteme extraction, which can be decomposed into, to be split to the camera lens of video and to the camera lens after segmentation Do image, semantic extraction.Block-based video lens cutting techniques, can be by the related shot segmentation of content together, Ran Houzai Choose and close the key frame in camera lens to represent the camera lens;Image, semantic extractive technique be video semanteme extraction committed step it One, main detection and the extraction of semantics of classification and destination object including to destination object.
SVM is a kind of learning model of SVMs, substantially a grader.It is solving small sample, non-thread Property and high dimensional pattern identification in show many distinctive advantages.SVM is a kind of learning model for having supervision of SVMs, Commonly used to carry out pattern-recognition, classification and regression analysis.A substantially grader.It is that linear is divisible Situation is analyzed, by using non-linear map that the low-dimensional input space is linear not in the case of linearly inseparable The sample that can divide, which is converted into high-dimensional feature space, makes its linear separability, so that high-dimensional feature space uses linear algorithm to sample This nonlinear characteristic carries out linear analysis and is possibly realized.
XML is a kind of important software engineering as a kind of extensible markup language.It can by it is a kind of it is flexible in a manner of The structure of management information, the structural relation having in itself with different node hierarchy description information.Dom4j is that dom4j.org goes out One of product increases income XML parsing bags, and user can read and write each node content of XML file with dom4j technologies.
The present invention solves destination object social networks identification problem using the technology such as Video processing, SVM, XML.
The content of the invention
Technical problem:It is an object of the invention to provide a kind of destination object social networks identification side based on video semanteme Method, Social Media resource contain abundant semantic information, and wherein video is to obtain social semantic important sources, but current main It is related to mark social semantic to identify to rely on people, lacks effective technology by software analysis semantic information to excavate video The social networks that middle destination object is contained, present invention aim to address this problem.
Technical scheme:The present invention pre-processes to video data first, by partitioning video data into a series of camera lens, Obtain the semantic collection of destination object in camera lens;Then, related camera lens is formed into language to sequential relationship according still further to the content of camera lens Adopted sequence, form specific scene;Finally, the social pass between destination object is excavated by analyzing the semantic information of scene System.
Destination object social networks recognition methods proposed by the present invention based on video semanteme comprises the steps of:
Step 1) pre-processes to the video data of user's input first, and specific handling process is as follows:
Step 1.1) is split using block-based comparative approach to video data, obtains the camera lens of the video data, The block-based comparative approach is the region unit that the image of each frame of video data is divided into user's specified quantity, passes through ratio Similitude compared with the region unit between successive frame marks off different camera lenses, and wherein frame is a least unit i.e. frame for video data Image, camera lens are one group of continuous frame sequences in video, and the feature and specific standards of the similitude of the region unit are referred to by user Fixed, the region unit between the successive frame of same camera lens has similitude;
That frame of extraction in the camera lens frame sequence centre position is used as pass to step 1.2) from each camera lens successively Key frame, the key frame represent the camera lens in subsequent treatment;
Step 2) extracts the semantic collection for the destination object that user specifies in all key frames, and semanteme collection is converted into key-value pair Form be saved in the file of XML format;The destination object includes background object and the class of foreground object two, foreground object Who object, background object are place residing for personage, temporal information;The semantic collection is that destination object extracts in video Semantic information set, comprising background, the time, dialogue, personage, color, shape, texture semanteme;Described XML format File includes 3 layers of nested node, and first layer is scene node, uses<scene>Labeled marker, described scene refer to according to camera lens Semantic information and camera lens between sequential relationship composition one group of shot sequence;The second layer is camera lens node, is used<short>Mark Label sign;Third layer is specific semantic node, is used<key>Labeled marker;Extract the target that user specifies in each key frame The specific handling process of the semantic collection of object is as follows:
Step 2.1) carries out the detection and classification of destination object to key frame, extracts all targets that the key frame includes Object, while record the dialog information in the key frame between personage and the key frame and be located at the time point of the broadcasting in video;
Step 2.2) extracts the visual signature of all foreground and background objects of key frame, forms corresponding characteristic vector, institute Stating the visual signature of background object includes color, texture;The visual signature of foreground object includes color, texture, shape;
Step 2.3) is learnt with SVM to the characteristic vector of destination object in key frame, extracts foreground object and the back of the body The semantic information of scape object;The semantic information of the foreground object is the semantic information of the visual behaviour performance of foreground object, is wrapped Include color, shape, texture, personage, dialogue;The semantic information that the background object takes is the semantic letter of environment residing for background object Breath, including background, time, the SVM are a kind of learning models for having supervision;
Step 2.4) is by the foreground object of the key frame of acquisition and the semantic information of background object, according to the form of key-value pair It is saved under the camera lens node of XML file;
Step 3) analyzing step 2) obtained node lower time of each camera lens and the semantic node corresponding to personage, The camera lens node for possessing identical personage's semanteme node is classified as one group of camera lens node;Described personage's semanteme node is exactly XML texts In part<short>Under node<key>Name attribute is that key-value pair of personage in node;
Step 4) being incremented by according to the nodal value of entitled time node by the data for the every group of camera lens node classified Order is saved under the scene node of XML file, constructs camera lens semantic sequence successively, represents scene one by one;
Step 5) parses each scene node in XML file successively, analyzes its all semantic information included, obtains The semantic information of relation and personage between personage, these information of each scene are saved in matrix one by one successively, this Every a line of a little matrixes or the element of each row store the relation between a personage and other personages and the semanteme of the personage Information, line number or row number of each personage in a matrix are specified by user;Described Scene Semantics information includes personage Between social networks and personage semantic information, wherein the semantic information of the social networks of personage and personage is saved in into one The specific handling process of matrix is as follows:
Step 5.1) extracts the semantic node in all camera lens nodes under XML file Scene node, obtains the scene All semantic informations;
Step 5.2) establishes a square according to this from the semantic information for obtaining finding out personage in all semantic informations of scene Gust, in matrix in addition to diagonal entry, when the line number of a row element is identical with a column element row number, this row element and one arranges Element represents the social networks of same personage, and cornerwise element preserves the semantic information of the personage;Cornerwise member The line number of element is identical with row number;
Step 5.3) is to the element progress assignment of scene homography, all semantic informations obtained from step 5.1), The social networks between personage and the semantic information of personage are extracted, then preserve the social networks of personage with set HashMap successively And semantic information, by element of the aggregate assignment to the correspondence position of matrix;Described set HashMap is one and is used for depositing key The data acquisition system of value pair;
Step 6) obtains a matrix for representing video semanteme information according to all matrixes for representing Scene Semantics;The matrix The semantic information and social networks of all persons in video is preserved, every a line of matrix or the element of each row store a personage The semantic information of relation and the personage between other personages, line number of each personage in a matrix or row number by User is specified, and each of which personage line number in a matrix or row number are specified by user, and idiographic flow is as follows:
Step 6.1) extracts all persons and the corresponding personage of each personage is semantic from the matrix of all Scene Semantics Information aggregate HashMap, successively by these semantic information collection conjunction unions, merge and be saved in a HashMap set, then will HashMap set after the merging is saved in the corresponding diagonal entry of matrix;
Step 6.2) extracts the social networks between personage from the matrix of all Scene Semantics, according to personage in matrix In line number or row number, successively by the social networks collection conjunction union of identical personage, merge and be saved in a HashMap collection Close, then the HashMap set after merging is saved in the position of each personage in a matrix.
Beneficial effect:The present invention first preserves video content structuring, is easy to computer to identify and analyze video semanteme, from And the social networks contained in video can be effectively inferred, widen the mode for excavating social networks.Specifically, this hair Method belonging to bright has following beneficial effect:
(1) present invention uses XML technology that the Content Transformation of video into structured data format, is easy to preserve and accumulate in video The inherent semantic information contained.A kind of architecture basics are provided to parse the semantic information contained in video and social networks below.
(2) present invention extracts from multi-angle to video semanteme, can obtain abundant semantic information, is accurate below The social networks contained in ground analysis video provide abundant content basis.
Brief description of the drawings
Fig. 1 is flow chart of the method for the present invention,
Fig. 2 is the Scene Semantics information matrix structure chart of the present invention.
Embodiment
Present invention specific implementation is described in more detail with reference to Fig. 1.
First, the semanteme of destination object in camera lens is extracted
Destination object detection and classification are carried out to camera lens first, foreground object and background object are extracted in camera lens, and Their vision and behavioural characteristic, time and human dialog information are recorded, and the characteristic vector of destination object is analyzed what is come Semanteme is saved in XML file in the form of key-value pair<short>Under node<key>In node.It is as follows:
Specifically start with when extracting vision and behavioural characteristic from color, texture and shape tripartite's region feature, divide in XML file Not with entitled color, shape and texture<key>The value of node represents the semantic information from this three aspects acquisition;Time For distinguishing different camera lenses, also the construction for semantic sequence below is given a clue, with entitled time's<key>Ode table Show;Goal object is primarily referred to as people, with entitled object's<key>Node represents;Dialogue in camera lens is with entitled Dialogue's<key>Node represents which destination object obj_name attribute representative this feature nodes belong in node.
XML format file is transformed into camera lens<short>During node, SVM models are used to train study Vision, behavioural characteristic vector inside camera lens sample, to complete the classification to destination object, then obtain with SVM models and remove sample The characteristic vector of camera lens in addition, and changed with conversion regime same above, the semantic information of each camera lens is saved in XML File<short>Under node.
2nd, scene is constructed according to shot sequence
Each scene is made up of one group of associated shot sequence.Come according to the contact between time series and destination object Construct shot sequence.And above all camera lenses are being converted in corresponding XML file<short>Node, so, structure It is exactly to parse this group with dom4j technologies to make scene<short>Node, pass through matching<short>In node file in key nodes Name attributes to be worth corresponding to object, find out and consistent with this value own<short>Node, it is incremental according to the time Order is saved in XML file successively<scene>Node gets off to represent this group of shot sequence i.e. video a scene.Following institute Show:
3rd, Scene Semantics are analyzed
Scene Semantics, substantially a n dimension square formation are built with a matrix, n is all prospects included in video Object number, this refers to personage's number.Above by the scene XML file in video<scene>Node represents, XML file is parsed with dom4j technologies on this basis, the content of parsing is saved in constructed matrix.Wherein square Each element of battle array is a HashMap set, and preservation is that the camera lens is corresponded in XML file<short>Key under node Value parses XML file to data, circulation<scene>Node information, all camera lens semantic informations of each scene are preserved Into corresponding HashMap.
4th, Scene Semantics matrix
Successively parse XML file in each scene node, analyze its all semantic information included, obtain personage it Between relation and personage semantic information.The semantic information of each scene is saved in matrix one by one successively.Described field Scape semantic information includes the social networks between personage and the semantic information of personage.Wherein by the social networks of personage and personage The specific handling process that semantic information is saved in a matrix is as follows:
(1) the semantic node in all camera lens nodes under XML file Scene node is parsed, obtains all semantic letters Breath.
(2) semantic information of personage is found out from obtained semantic information, establishes a matrix according to this.Except diagonal in matrix Line number and row number identical that two constituent element element beyond line element, represents the social networks of same person, and cornerwise element is protected Deposit the semantic information of corresponding personage.
(3) assignment is carried out to the element of matrix, analyzes obtained all semantic informations, obtain the social pass between personage System and the semantic information of personage, then gather the social networks and semantic information for preserving corresponding personage with HashMap successively.Described HashMap set is a data acquisition system for being used for depositing key-value pair.Finally successively by correspondence position of the aggregate assignment to matrix Element.
5th, video semanteme matrix
A matrix for representing video semanteme information is obtained according to all matrixes for representing Scene Semantics.The matrix is preserved and regarded The semantic information and social networks of all persons in frequency.Every a line of matrix or the element of each row store a personage and other The semantic information of relation and the personage between personage, line number or row number of each personage in a matrix are referred to by user It is fixed.Each of which personage line number in a matrix or row number are specified by user.Idiographic flow is as follows:
(1) from the matrix of all Scene Semantics, all persons and the corresponding personage's semantic information of each personage are extracted Set HashMap.Successively by these semantic information collection conjunction unions, a big set is merged into, then the big set is protected It is stored in the corresponding diagonal entry of matrix.
(2) the social networks set HashMap between personage is extracted from the matrix of all Scene Semantics.According to personage Classification, successively by the social networks collection conjunction union of identical personage, is merged into a big set.According to personage in a matrix Line number or row number, the big collections of the social networks of each personage is preserved into corresponding position in a matrix.
Embodiment is expanded on further with reference to case.
Assuming that there is one section of video, describe there are four people waiting car in bus platform, represent four with A, B, C, D respectively Personage, video only include two scenes.
(1), scene one:A laughs at has said sentence to B, " sees a film together at night!", then B, which laughs at, answers:" it is dear, You are happy, and what is all right.”
(2), scene two:C and D look at the mobile phone of oneself, between there is no any exchange.
The scene has four target persons, so with semantic, four human targets A, B, C, D between 4 dimension square formations storages Represent.The first row storage A of square formation and the semantic information occurred between other people, below by that analogy.Each element of square formation Value, storage be expert at corresponding to target person with the corresponding target person of row, the semantic information extracted in video.
Scene one includes two shot sequences:
1) A laughs at has said sentence to B, " sees a film together at night!”
2) B, which laughs at, answers:" dear, you are happy, and what is all right.”
Scene two includes a camera lens:1) C and D is seeing the mobile phone, and never exchanges.
It is as follows to be converted into XML file for camera lens in scene one:
Scene Semantics information matrix structure is as shown in Figure 2.
Social networks are analyzed:
(1) A and B relations are very close.
(2) C and D relations are strange.

Claims (1)

  1. A kind of 1. destination object social networks recognition methods based on video semanteme, it is characterised in that the step of this method includes For:
    Step 1) pre-processes to the video data of user's input first, and specific handling process is as follows:
    Step 1.1) is split using block-based comparative approach to video data, obtains the camera lens of the video data, described Block-based comparative approach is the region unit that the image of each frame of video data is divided into user's specified quantity, is connected by comparing The similitude of region unit between continuous frame marks off different camera lenses, and wherein frame is a least unit i.e. frame figure for video data Picture, camera lens are one group of continuous frame sequences in video, and the feature and specific standards of the similitude of the region unit are specified by user, Region unit between the successive frame of same camera lens has similitude;
    Step 1.2) successively from each camera lens that frame of extraction in the camera lens frame sequence centre position as key frame, The key frame represents the camera lens in subsequent treatment;
    Step 2) extracts the semantic collection for the destination object that user specifies in all key frames, and semantic collection is converted into the shape of key-value pair Formula is saved in the file of XML format;It is personage that the destination object, which includes background object and the class of foreground object two, foreground object, Object, background object are place residing for personage, temporal information;The semantic collection is the language that destination object extracts in video The set of adopted information, comprising background, the time, dialogue, personage, color, shape, texture semanteme;The file of described XML format Comprising 3 layers of nested node, first layer is scene node, is used<scene>Labeled marker, described scene refer to the language according to camera lens One group of shot sequence of the sequential relationship composition between adopted information and camera lens;The second layer is camera lens node, is used<short>Label mark Show;Third layer is specific semantic node, is used<key>Labeled marker;Extract the destination object that user specifies in each key frame Semantic collection specific handling process it is as follows:
    Step 2.1) carries out the detection and classification of destination object to key frame, extracts all targets pair that the key frame includes As, while record the dialog information in the key frame between personage and the key frame and be located at the time point of the broadcasting in video;
    Step 2.2) extracts the visual signature of all foreground and background objects of key frame, forms corresponding characteristic vector, the back of the body The visual signature of scape object includes color, texture;The visual signature of foreground object includes color, texture, shape;
    Step 2.3) is learnt with SVM to the characteristic vector of destination object in key frame, extracts foreground object and background pair The semantic information of elephant;The semantic information of the foreground object is the semantic information of the visual behaviour performance of foreground object, including face Color, shape, texture, personage, dialogue;The semantic information that the background object takes is the environment semantic information residing for background object, Including background, time, the SVM is a kind of learning model for having supervision;
    Step 2.4) preserves the foreground object of the key frame of acquisition and the semantic information of background object according to the form of key-value pair To under the camera lens node of XML file;
    Step 3) analyzing step 2) obtained node lower time of each camera lens and the semantic node corresponding to personage, possessing The camera lens node of identical personage's semanteme node is classified as one group of camera lens node;Described personage's semanteme node is exactly in XML file< short>Under node<key>Name attribute is that key-value pair of personage in node;
    Incremental order of the step 4) by the data for the every group of camera lens node classified according to the nodal value of entitled time node It is saved under the scene node of XML file, constructs camera lens semantic sequence successively, represents scene one by one;
    Step 5) parses each scene node in XML file successively, analyzes its all semantic information included, obtains personage Between relation and personage semantic information, these information of each scene are saved in matrix one by one successively, these squares Every a line of battle array or the element of each row store the relation between a personage and other personages and the semantic information of the personage, Line number or row number of each personage in a matrix are specified by user;Described Scene Semantics information is included between personage Social networks and the semantic information of personage, wherein the semantic information of the social networks of personage and personage is saved in into matrix Specific handling process is as follows:
    Step 5.1) extracts the semantic node in all camera lens nodes under XML file Scene node, obtains the scene and owns Semantic information;
    Step 5.2) establishes a matrix, square according to this from the semantic information for obtaining finding out personage in all semantic informations of scene In battle array in addition to diagonal entry, when the line number of a row element is identical with a column element row number, this row element and a column element The social networks of same personage are represented, cornerwise element preserves the semantic information of the personage;Cornerwise element Line number is identical with row number;
    Step 5.3) carries out assignment to the element of scene homography, all semantic informations obtained from step 5.1), extracts The semantic information of social networks and personage between personage, then the social networks and language of personage are preserved with set HashMap successively Adopted information, by element of the aggregate assignment to the correspondence position of matrix;Described set HashMap is one and is used for depositing key-value pair Data acquisition system;
    Step 6) obtains a matrix for representing video semanteme information according to all matrixes for representing Scene Semantics;The matrix preserves The semantic information and social networks of all persons in video, every a line of matrix or the element of each row store a personage and its The semantic information of relation and the personage between his personage, line number or row number of each personage in a matrix are by user Specify, each of which personage line number in a matrix or row number are specified by user, and idiographic flow is as follows:
    Step 6.1) extracts all persons and the corresponding personage's semantic information of each personage from the matrix of all Scene Semantics Set HashMap, successively by these semantic information collection conjunction unions, merge and be saved in a HashMap set, then this is closed HashMap set after and is saved in the corresponding diagonal entry of matrix;
    Step 6.2) extracts the social networks between personage from the matrix of all Scene Semantics, according to personage in a matrix Line number or row number, successively by the social networks collection conjunction union of identical personage, merge and be saved in a HashMap set, The HashMap set after merging is saved in the position of each personage in a matrix again.
CN201510137760.1A 2015-03-26 2015-03-26 A kind of destination object social networks recognition methods based on video semanteme Expired - Fee Related CN104778224B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510137760.1A CN104778224B (en) 2015-03-26 2015-03-26 A kind of destination object social networks recognition methods based on video semanteme

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510137760.1A CN104778224B (en) 2015-03-26 2015-03-26 A kind of destination object social networks recognition methods based on video semanteme

Publications (2)

Publication Number Publication Date
CN104778224A CN104778224A (en) 2015-07-15
CN104778224B true CN104778224B (en) 2017-11-14

Family

ID=53619688

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510137760.1A Expired - Fee Related CN104778224B (en) 2015-03-26 2015-03-26 A kind of destination object social networks recognition methods based on video semanteme

Country Status (1)

Country Link
CN (1) CN104778224B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106294783A (en) * 2016-08-12 2017-01-04 乐视控股(北京)有限公司 A kind of video recommendation method and device
CN106778537B (en) * 2016-11-28 2021-02-02 中国科学院心理研究所 Animal social network structure acquisition and analysis system and method based on image processing
CN107909038A (en) * 2017-11-16 2018-04-13 北京邮电大学 A kind of social networks disaggregated model training method, device, electronic equipment and medium
CN107992598A (en) * 2017-12-13 2018-05-04 北京航空航天大学 A kind of method that colony's social networks excavation is carried out based on video data
CN108388836A (en) * 2018-01-25 2018-08-10 北京览科技有限公司 A kind of method and apparatus for obtaining video semanteme information
CN109471959A (en) * 2018-06-15 2019-03-15 中山大学 Personage's social relationships discrimination method and system in image based on figure inference pattern
CN109344285B (en) * 2018-09-11 2020-08-07 武汉魅瞳科技有限公司 Monitoring-oriented video map construction and mining method and equipment
CN110070438B (en) * 2019-04-25 2021-09-03 上海掌门科技有限公司 Credit score calculation method, device and storage medium
CN112132118B (en) * 2020-11-23 2021-03-12 北京世纪好未来教育科技有限公司 Character relation recognition method and device, electronic equipment and computer storage medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663095A (en) * 2012-04-11 2012-09-12 北京中科希望软件股份有限公司 Method and system for carrying out semantic description on audio and video contents

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7818667B2 (en) * 2005-05-03 2010-10-19 Tv Works Llc Verification of semantic constraints in multimedia data and in its announcement, signaling and interchange

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663095A (en) * 2012-04-11 2012-09-12 北京中科希望软件股份有限公司 Method and system for carrying out semantic description on audio and video contents

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Video semantic content analysis based on ontology;Bai L等;《IEEE》;20070924;全文 *
基于MPEG_7的视频语义描述方法;朱华宇等;《南京大学学报(自然科学)》;20020131;第38卷(第1期);全文 *

Also Published As

Publication number Publication date
CN104778224A (en) 2015-07-15

Similar Documents

Publication Publication Date Title
CN104778224B (en) A kind of destination object social networks recognition methods based on video semanteme
CN104679863B (en) It is a kind of based on deep learning to scheme to search drawing method and system
CN104063683B (en) Expression input method and device based on face identification
CN105740758A (en) Internet video face recognition method based on deep learning
CN105574510A (en) Gait identification method and device
Din et al. Service orchestration of optimizing continuous features in industrial surveillance using big data based fog-enabled internet of things
Tan et al. Attention-based pedestrian attribute analysis
CN105046720B (en) The behavior dividing method represented based on human body motion capture data character string
CN107729363B (en) Bird population identification analysis method based on GoogLeNet network model
CN107169106A (en) Video retrieval method, device, storage medium and processor
CN106649663A (en) Video copy detection method based on compact video representation
Wang et al. MetaSearch: incremental product search via deep meta-learning
Zhang et al. Online modeling of esthetic communities using deep perception graph analytics
Lwowski et al. Geospatial event detection by grouping emotion contagion in social media
Chen et al. A saliency map fusion method based on weighted DS evidence theory
CN107273478A (en) A kind of semi-supervised hashing image searching method based on Group Lasso
Liu et al. A general framework for co-training and its applications
Singh et al. Generative scatternet hybrid deep learning (g-shdl) network with structural priors for semantic image segmentation
CN102663445A (en) Image understanding system based on layered temporal memory algorithm and image understanding method thereof
CN111368943B (en) Method and device for identifying object in image, storage medium and electronic device
CN112417289A (en) Information intelligent recommendation method based on deep clustering
Luo et al. An optimization framework of video advertising: using deep learning algorithm based on global image information
WU et al. Knowledge representation of entity attribute frame for natural language understanding
CN103530656B (en) Hidden structure learning-based image digest generation method
CN108280163B (en) Video feature learning method and device, electronic equipment and readable storage medium

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
EXSB Decision made by sipo to initiate substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20181114

Address after: Room 929, 102 Chaoyang North Road, Chaoyang District, Beijing 100123

Co-patentee after: Li Bo

Patentee after: Fan Liyang

Address before: 210046 9 Wen Yuan Road, Ya Dong new town, Qixia District, Nanjing, Jiangsu.

Patentee before: Nanjing Post & Telecommunication Univ.

CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20171114

Termination date: 20190326