Two Systems for Temporal Video Segmentation

This paper presents two different temporal video segmentation systems and their performance evaluation using a standard test corpus. The first system uses two separate detection subsystems respectively based on color histogram comparison and on rough edge tracking, and then fuses their results. The second system uses direct image comparison after motion compensation. Both systems are described and evaluated using the ''AIM'' corpus and reference segmentation. Though very different in principle, both systems perform similarly well on the test database. Combining the systems' output even in very simple ways yields a significant performance improvement.