Font Size: a A A

Video Feature Storage Technology Application Research Based On DCI Video Works

Posted on:2014-01-02Degree:MasterType:Thesis
Country:ChinaCandidate:L YangFull Text:PDF
GTID:2248330395498589Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
In the digital work registration service system, work and work feature storage i s the foundation work of registration. Because video works have data characteristics of different types, large amount, complicated structure, the storage and query effici ency directly affects the practicability of registration platform, the research has imp ortant significance. Storage technologies include video works storage and video feat ure storage, the research related to feature extraction, feature indexing, feature stora ge and other issues, which are the core problem for video registration.Based on the analysis of the present domestic and foreign video storage and re trieval technology, this paper proposed a video hierarchically structured data model for the storage technology of video works. For the key frame extraction, this paper proposed an improved mutual information key frame extraction algorithm to overco me the difficulties of threshold selection and algorithm complexity present in previo us key frame extraction methods. It constructed a mutual information similarity curv e detection algorithm based on a non-uniform, weighted HSV histogram. For the lar ge amount of calculation and the error matching problems in points feature extracti on, this paper proposed method that extract SIFT feature points based on global inf ormation. To solve the high dimension problem, this paper proposed algorithm that combine global information of SIFT features with LSH high-dimensional cluster ind ex algorithm. To solve the inaccurate retrieval effect problem existed in the features signature distance alignment algorithm, this paper proposed another retrieval metho d of comprehensive sequence feature curve combined with the key frames features of video.In this paper, the result was tested by comparing the same algorithm for differ ent types of video as well as comparing the same type of video for different algori thm. The result was evaluated by recall and precision. Experimental results show th at the algorithm can do automatic key frames extraction without threshold selection. In addition, this method also extracts key frames quickly and accurately after only a single scan. The retrieval method that combined the video sequence and key fra mes features have a very good retrieval effect on various types of video and have a high recall and precision.
Keywords/Search Tags:Copyright Protection, Feature Extraction, Feature Storage, FeatureIndex, Similarity
PDF Full Text Request
Related items