Color And Shape Features Based Video Retrieval Research

With the rapid development of network technology and increasement in the amount of data of multimedia information, video as multimedia information in the most complex kind of type of media, has been widely applied and developed. How to quickly and accurately find video library from huge video clips of interest video has become the key problem in the development of information field. Content_Based Video Retrieval is a retrieval process of video data from database of large-scale video according to its context.On the base of Content_Based Video Retrieval research, boundary detection, key frame extraction, and feature extraction are made further research and improvement. The main research work is as follows:Firstly three key technologies and video retrieval framework are introduced, and video retrieval process is known.Secondly lens boundary detection and key frames extraction are deeply researched in the video retrieval. Lens boundary detection is divided into shear detection and gradual change detection. Various detection methods are introduced, and dual window method is used to achieve lens of shear test. Average-based method and threshold-based method are used to extract key frames on camera in the key frames extraction.Thirdly content-based image retrieval is briefly introduced, and retrieval algorithms based on color features, shape features and comprehensive features, are achieved. An improved model based on HSV local cumulative histogram is proposed. Fourier descriptor is used to describe contour feature in shape features-based retrieval algorithm. Then the two features are normalized similarly and summed and averaged to retrieve images based on comprehensive features. The best weight is (0.6,0.4), which is got by repeated experiments.Finally a general video retrieval system based on content is constructed, in which videos can be retrieved quickly according to video content and text description. There are some shooting video clips, movie clips, MV clips and advertising clips. The experimental results show that the retrieval rate of the proposed algorithm is higher than traditional method.