BUPT at TRECVID 2007: Shot Boundary Detection

In this paper we describe our methodologies and evaluation results for the shot boundary detection at TRECVID 2007. We submitted 10 runs results based on SVM classifiers and several separate detectors. BUPT_01 Default SVM parameters and a low threshold for motion detector BUPT_02 Default SVM parameters and a low threshold for edge detector BUPT_03 Make high penalty for false cuts to increase the precision BUPT_04 Make more penalty for false cuts to increase the precision BUPT_05 Make high penalty for false gradual transitions (GT) to increase precision BUPT_06 Make high penalty for missing GT to increase recall BUPT_07 Enhance motion threshold to increase the precision of GT BUPT_08 Extend the filtering window of motion to increase both recall and precision of GT BUPT_09 Increase the recall of both cuts and GT BUPT_10 decrease motion threshold to increase the recall of GT Evaluation results showed that our CUT algorithm can achieve a satisfying result while the GT dose not work as well as our previous testing.