Video Scene Segmentation Based On Evolutionary Algorithm

With the development of computer video technology, it has become a focus of computer vision research to make the computer understand the video's content. Scene segmentation is the first step toward semantic understanding of video, and it also provides the effective approach to browse and index video database. Video Scene segmentation is done at a higher level, it bases on the result of video shot detection and key frames abstraction, integrates the professional knowledge of scene detection. Shot-detection is the precondition of scene segmentation, on the basis of full investigation on current digital video and video segmentation technologies, this paper proposed a shot detection method which based on HSV color model; In computer scene detection, this paper applies the evolutionary algorithm to the video scene segmentation field, and forms a new scene detection method which is completely different from the current.The shot detection method proposed by this paper transforms the video pictures from RGB space to HSV space firstly. Then requantitate the H and S components of pictures to 127 color areas to overcome the interference produced by illumination changing and movement. Lastly compute the similarity between adjacent frames by histogram. Further more, it adopted the false-shot-analysis approach to weaken the interference brought by fast change of pictures. The experiment results show that this method has higher accuracy and recall ratio, as well as faster processing speed. And it has the strong practicability in shot detection application. it cooperates well with the scene segmentation method proposed by this paper to segment the video scene.In essence, Evolutionary algorithm is a kind of direct search method which do not depend on the specific problem, it improves the individuals' adaptability by the mechanisms such as natural selection, inheritance, mutation, etc. It is a brand-new attempt to apply the evolutionary algorithm to the video scene segmentation field. In the process of scene segmentation, this paper views the division points of shots as genes, uses the special gene coding mode, populations evolvement regulations and individual evaluating mechanism. The experiment results show that it is a simple but efficient video scene segmentation method, and it has higher feasibility of scenesegmentation.