Abrupt and gradual scene transition detection in MPEG-4 compressed video sequences using texture and macroblock information

In video indexing, video scene transition detection is an essential step to segment the video. Current techniques used for scene change detection tend to suffer from a major limitation as most of them cannot identify scene transitions in the compressed domain. Since most videos are expected to be stored in the compressed domain, scene transition detection in this domain is highly desirable. In this paper, we propose an algorithm for real-time detection of abrupt scene transitions and fade-in/fade-out scene transitions using texture and macroblock (MB) information in the MPEG-4 compressed video. MPEG-4 video stream is partially decoded where texture and MB information are extracted. MB information is used to identity the abrupt scene transitions whereas MB information together with the texture difference between frames are used to identify the boundaries of fading transitions. The performance of the proposed algorithm is measured in terms of precision and recall values. The experimental results show that the proposed algorithm can identify fade and abrupt scene transitions with a mean recall and precision values equal to 80.76% and 85.71%, respectively.