Abstract
A mask matching approach to detect shot changes in MPEG coded video uses reference ratio variances of macroblocks between MPEG coded frames. A function is designed to quantize the results into shot change probability values. Moreover, a conversion function modifies this probability computation to minimize misdetection and loss of detection under unusual image pattern situations. A shot change probability threshold is defined for the video to be examined. When a frame's modified shot change probability exceeds this threshold, the frame is considered to contain a shot change. With this approach, processing time is reduced by evaluating MPEG coded data directly, rather than the raw video.