| As a kind of news information, news video information plays an important role inthe fields of politics, economy, culture and daily life. In order to help users obtain keysemantic information and get new interested or track information from massive videodata, more and more scholars pay their attention to news video semantic retrieval andits automated processing in recent years.Started from analysis of the key semantic news video, this dissertation mainly didsome in-depth study such as the construct processing of news video, extraction ofsemantic feature and automatic classification, and got some innovative achievementsas below:In order to understand and analysis the news video information better, this issueanalyzes the relationship between multimedia elements and the six key elements of thenews (that is: When, Where, Who, What, according to, and How) in news videoinformation, put forward the method of news video semantic extraction which arebased on the news of the key elements and proposed the model of news video semanticanalysis which is based on key press elements.What’s more, in order to extract the semantic information of the news videoscenes, characters, and events, this subject studied the methods of news video structureanalysis and critical semantic feature extraction, achieved the extraction of news videosemantic features from video key frame, visual features, text features and specifictechnologies such as color histogram, color moment, Canny edge LBP texture features,SIFT, CRF segmentation system and so on.At last, this issue brought forward an auto method to classify news video basedon hierarchical conditional random fields in order to realize the classification of newsvideo automatically. And this method utilized the methodology which mixed the low-level visual features and text features of the news video semantic to achieve the goal ofauto classification. Results of tests by training and testing some typical kinds of new videos fromyouku.com verified the validity of methods proposed in this dissertation. Adoptingmethods mentioned above could more do auto classification accurately, which couldreduce the workload of website editors greatly and make it easy for users to track thedevelopment trend of similar news at the same time. |
PDF Full Text Request