The SAMOVA Shot Boundary Detection for TRECVID Evaluation 2004

This paper describes the system used by the SAMOVA team (IRIT) to perform the shot boundary detection task of the TRECVID 2004 workshop. Our main algorithm is based on detection of illumination changes, as well as on the length of these transitions. The results are then refined by ckecking all the candidate transitions. We propose three versions of our application. Samova01: the simplest form, only uses illumination changes with image difference. Samova02: checks every cut, using morphological information. Samova03: checks both cuts and gradual transitions. The second and third runs are improved versions of the first run. They use a small threshold to have more detections, but add a validation test for each detected transition. Our application is made to work with compositing effects, which are editing effects. So, the evaluation of our application makes sense if we take into consideration compositing effects.On video contents where such effects are representative of the editing work our results are improved by the ability of our methods to deal with effects such as overlay, captions suddenly appearing on the screen, . . .