Indian Institute of Technology , Bombay at TRECVID 2005

This year, Indian Institute of Technology, Bombay participated in TRECVID 2005 in the task of shot detection. We observe that, even though large number of available shot detection methods perform well under normal conditions (as evinced by the results at Trecvid for the last couple of years), they fail when there are deliberate or inadvertent lighting variations. This might result in labeling some frames as shotbreak, where there is no actual shot-break (false positives). We have developed a method to reduce the number of such false positives significantly. Our primary focus is on hard cuts. Our method works in two passes. The first pass predicts approximate locations of shot-breaks in the video sequence. This is done using wavelet analysis of correlation values between successive frames in a video sequence. The second pass consists of detection and removal of false positives from predicted shot breaks of the first pass.