Research On Content Based Video Retrieval Algorithms

With the fly of time and rapid development of technology, video data expands greatly. So how to retrieve Video data is becoming an important problem. In order to effectively retrieve video data, video data must be effectively organized, indexed, if so, users can get what users want in short time. So content based retrieval, or CBR, which is on the basis of the objective characts of video data, is an effective solution to video retrieval and the focus in multimedia retrieval field at present.In video data retrieval, characteristics extraction is primary problem. In order to give response as soon as possible, improve the algorithm of extraction of color characteristics. The algorithm reduces the cost of computation, but can represent color characteristic, and give a more reasonable result of extraction of color characteristic. When extracting the feature of object, recognizing a object in video is firstly needed, so give a algorithm of edge detection based on wavelet.Video retrieval is to find relevant videos by computing in high-dimensional space, the key is how to speed up the retrieval. In order to get over dimension curse problem, A feasible method is used. The method is to construct a Kernel Vector, filter irrelevant data on each dimension of Kernel Vector and reduce the search range.In video data retrieval, relevant feedback is adopted to improve the veracity of retrieval by adjusting weights of every sort of characteristics, and finally approach the users'perception about videos.In the experimental system, some parts of shot segmentation,key frame extraction, static character extraction and video data retrieval are achieved. By testing, the correction and veracity of the system are validated.