Cut detection method for MPEG compressed video using likelihood ratio test

✍ Scribed by Toshimitsu Kaneko; Osamu Hori

• Publisher: John Wiley and Sons
• Year: 2000
• Tongue: English
• Weight: 220 KB
• Volume: 31
• Category: Article
• ISSN: 0882-1666
• DOI: 10.1002/(sici)1520-684x(200006)31:6<19::aid-scj3>3.0.co;2-4

✦ Synopsis

In this paper, we propose a cut detection method, which is the most basic process in video analysis, based on likelihood ratio testing for MPEG compressed video. This method detects cut points by learning the likelihood of the presence or absence of a cut point from samples by observing the differential motion vectors in the MPEG bitstream and applying statistical hypothesis testing. Thus, the threshold which had been determined from heuristics can be set analytically. The test results based on 250 minutes of video for training and 250 minutes of video for evaluation demonstrated the superior detection performance of the proposed method (a 12% false negative rate and a 5% false positive rate) over the conventional method of counting specific types of macroblocks and showed that differential motion vectors are more suitable observation information than DCT dc components.