Analysis And Research Of Content-Based Video Retrieval Techniques

With the emergence of huge amounts of multimedia data and further research onInformation Retrieval, the traditional retrieval methods based on the text can no longer besatisfied by the needs of applications and Content-Based Video Retrieval(CBVR) has beingfocused which could retrieve information by commands through these mass data accuratelyand rapidly and through their feature matching based on their own contents. However,obstacles of CBVR are how to implement the organization, expression and indexing of thevideo data and to realize the mapping between underlying features and high-level semanticsand to accomplish in networks for the complex structure of video data.Based on the relevant background and technology of video retrieval, studies on thevideo indexing, semantic mapping and access to online video packages deeply in the thesiswhich include the followings:1. Provided a hierarchical structure based on high-level semantics after the analysis ofvideo data structure which connects the underlying features with high-level semantics and itcan considerably reduce searching range and improve the efficiency through a collaboratingretrieval between multiple views in retrieval.2. With the hierarchical structure, the retrieving contents are divided into multi-objectinformation frames and single-object information frames. After comparing these frames withthe high-level semantic library, the video indexing structure by constructing a multi-objectindex and a single-object index and combining are improved.3. To the better performance in clarification and relevance feedback, SVM-basedrelevance feedback algorithm is improved with the experiments show that the advanced thealgorithm can make a better use of limited feedback in video data labeling.4. Introduce an online video package extraction algorithm integrating linear hash functionand XOR displacement operation by making full use of the characteristic of interaction ofonline video stream with experiments show that the algorithm has better time performance.5. On the basis of preliminary work of this paper, a CBVR system framework on sportsvideos had been designed and proved that this system performs highly efficiently inexperiments.