一种新的镜头检测算法A New Shot Boundary Detection Algorithm.pdfVIP

A New Shot Boundary Detection Algorithm Dong Zhang, Wei Qi, Hong Jiang Zhang Microsoft Research, China hjzhang@ Abstract. Shot is often used as basic unit for both analyzing and indexing video. In this paper, we present an algorithm for automatic shot detection. In our algorithm, we use a ‘flash model’ and a ‘cut model’ to deal with the false detection due to flashing lights. A technique for determining the threshold that uses the local window based method combined with reliability verify process is also developed. The experimental results of our method are very promising, improving the performance of shot detection. 1 Introduction There are many shot detection methods already proposed in past decades [1-6]. The common way for shot detection is to evaluate the difference between consecutive frames represented by a given feature. Although reasonable accuracy can be achieved, there are still problems that limit the robustness of these algorithms. One of the common problems in robust shot detection results from the fact that there are many flashlights in news video, which often introduce false detection of shot boundaries. Only some simple solutions to this problem have been proposed in [2]. Their main limitations are that they assume the flashlight just occur during one frame. In real world, such as news video, there will be many flashlights occur during a period of time and influence multiple consecutive frames. Another problem that has not been solved very effectively well is threshold selection when comparing changes between two frames. Most o