Research And Application On Shot Boundary Detection And Key Frame Extraction

With the rapid development of computer and multimedia technology, contented-based video retrival (CBVR) has become one of the hottest research topics. This thesis deals with shot boundary detection and key frame extraction, which are two main tasks of content-based video retrival.Based on the analysis of existing shot cut detection methods, the author proposes a novel, fast cut detection algorithm, which is based on the combination of string matching and histogram intersection. Experiments show that this method can alleviate the false detection rate caused by using only string matching method while keeping satisfactory recall rate with high precision. As gradual shot transition detection is concerned, this thesis attempts to combine several kinds of image features, such as histogram, moments and edges, and then support vector machine is employed to detect gradual shot transition. Our experiments show that this method can detect video gradual shot transition, while the precision still needs improvements.At last, we implement a content based video retrieval prototype system which combine video shot transition detection and key frame extraction. People could find the program of interest, and then can browse it. During the implementation, some improvement has been made to key frame extraction by taking the content variation into consideration.