An Improved Moving-Window Method for Shot Boundary Detection

Shot boundary detection is an important phase in the process of video retrieval, thus an automatic way to do this work is imperative. In this paper, we present an improved moving-window approach which can tackle this problem effectively. In our method, an existing concept of moving window is adopted. Then the Manhattan distance of histograms in RGB color space and the Laplace transform of each frame are calculated as the measurement of similarity. And finally each frame is examined through several classifiers to determine whether it is a component of a transition. The innovation of this paper is to apply Laplace transform as a supplementation of current features, and to propose a flexible video-dependent threshold, which leads to a higher precision. Experiments show that our approach can generate preferable results in most cases compared with two previous methods.