Video Shot Detection Combining Multiple Visual Features

1 Overview Shot based indexing techniques have been widely used to organize video data. Scene change detection is the most commonly used method to segment image sequences into coherent units for video indexing. A shot is a sequence of contiguous frames that are recorded from a camera. There is usually one continuous action within a shot, with no major change of scene content. However, there are still many different changes in a video (e.g. object motion, lighting change and camera motion), it is a nontrivial task to accurately detect scene changes. Furthermore, the cinematic techniques used between scenes, such as dissolves, fades and wipes, produce gradual scene changes that are harder to detect. Scene cut detection algorithms have been studied since the early 90's. The basic method is to measure the pixel difference frame-to-frame in terms of intensity or color [8]. In [8], the number of changed pixels is counted and if the number exceeds a certain percentage, a scene cut is detected. This method is not robust due to the camera and object motions that can cause large pixel value differences.