A Unified Framework for Shot Boundary Detection

With the development of computer and TV technology, the content-based video retrieval system becomes more and more important to organize, index and retrieve the massive video information in many application domains. Automatic shot boundary detection is a crucial issue in content-based video retrieval and has attracted attention of many researchers in recent years. However, few literatures are reported to make a unified method to detect the types of shot boundaries, such as cut, fide in/out and dissolve. In this paper, a unified framework is proposed to identify the various shot boundary types. Firstly, each video frame is represented as a point in HS color space, and the video sequence is mapped into a normalized feature curve by using the L2-norm. The feature curve can hold shot boundary characteristic very well. Then the detection of singularities with wavelet is employed to identify the potential shot boundaries in the feature curve. At last, the shot boundaries are further classified according to the characteristic of normalized feature curve. The experimental results compared with the conventional schemes achieve satisfying precision and recall of detecting shot boundaries in this scheme.