Study Of The Methods For Shot Boundary Detection Of Video Sequences

With the rapid development of multimedia technology and widespread use of Internet,more and more dada and information can be obtained easily. How to organize, manage and indexthese data and information leads to a new research field of video processing, i.e.,content-based video retrieval and indexing. The main task of content-based video retrievaland indexing is the shot boundary segmentation. Now, automatic shot transition detectionis still an open and challenging issue, which is emphasized in this paper.Firstly, Based on the analysis of existing techniques, a new idea of applying themultiresolution of wavelet transforms to shot transition detection is proposed. Withinvestigation and analysis of various kinds of shot transition, a joint detection schemeis presented with an integration of different detection techniques. Secondly, accordingto both the large data of video and the features of various existing methods, a newidea of cutting before detection is proposed, and the schemes of the un-blind and blindadjustment of detection character or function are given. The experimental results onactual video data demonstrate that the proposed approach is of good precision andgenerality. Thirdly, under the consideration of the first generation wavelet's heavycomputational burden, an integer-to-integer wavelet is applied into the field of shotboundary detection. The dissimilarity function is defined by norms theory. Takingaccount to the difficulty of threshold determination and the easy confusion of gradualtransition with motion, an automatic threshold selection method and a scheme oftwice search by Tchebycheff distance are presented. Simulation results of differentvideo sequences demonstrate that the proposed method is more effective than thetraditional ones. Finally, with an introduction of There Dimensional WaveletTransform to the shot detection and making use of its spatial-temporal relativity, amethod based on motion vector is given, which is more efficient than others.The detection schemes and methods proposed in this paper have the ability bothto detect shot and to identify and locate the boundary, with excellent performancesover the traditional techniques.