Illumination Invariant Shot Boundary Detection

Illumination variation poses a serious problem in video shot detection. It causes false cuts in many shot detection algorithms. A new illumination invariant measure metric is proposed in this paper. The metric is based on the assumption: the outputs of derivative filters to log-illumination are sparse. Thus the outputs of derivative filters to log-image are mainly caused by the scene itself. If the total output is larger than a threshold, it can be declared as a scene change or a shot boundary. Although this metric can detect gradual transitions as well as cuts, it is applied as a post-process procedure for a cut candidate because an illumination change is usually declared as a false cut.