Reliable real time scene change detection in MPEG compressed video

This work presents a new scene change detection method in MPEG compressed video. This method is based on two different video characteristics, edges and luminance, and takes the best of them to define two distance functions. On one hand, a contour based distance function that requires no registration technique is introduced. On the other hand, a new distance function based on luminance histograms correlation is defined. In order to efficiently combine both distance functions, a new estimator based on the divergence of distributions is proposed. The method has been tested using several videos from the MPEG-7 content set.