Combining luminance and edge based metrics for robust temporal video segmentation

This work presents a new video indexing technique able to detect shot cuts as well as gradual transitions in MPEG compressed video sequences. Moreover, this technique also allows to perform global motion estimation in order to successfully detect swing, zoom and displacement effects in video streams. In fact, motion information is used during the gradual shot transition detection algorithm in order to reduce false positives caused by video motion. Two different metrics are used. The first one is based on luminance histograms, while the second is based on the Generalized Hough transform. The use of this technique greatly improves the results obtained by other shot cut detection algorithms. Also, it is a key technique to the success of both; the gradual shot transition detection and the global motion estimation algorithms.