Jdl at Trecvid 2006 Shot Boundary Detection

In this paper we proposed a simple but effective method for simultaneously detecting video shot transitions of various types by means of combining threshold and SVM based method. Our method first selects the suspicious transition candidates using a low threshold method and then judges the candidates by using the SVM base method. In selecting the low threshold, we use only two simply features: histogram and mutual information. Due to this low threshold, we do not need to extract complex features around every frame changes, so our method can run over real time. The most important problem of the system is its fail to distinct gradual transition and motion in some times. This is because we use little information about the camera motion and object motion. In the future we will use more information about the camera motion to alleviate this problem. The test result on TRECVID benchmark shows our approach is effective for shot boundary detection.