Abstract: Shot segmentation is structural base for video retrieval. An efficient shot segmentation algorithm was proposed in this paper. Firstly aiming at the motion among the video frames,motion vector field adaptive search technique was used to estimate block motion vectors between two video frames and then motion compensation was made. Afterwards on this base the number of pixsels whose value don't change much between two frames were computed and then based on sliding window, local self-adaptive dual thresholds were gained. Finally based on comparison between dual thresholds, cuts and gradual transition were detected. Experimental results prove that this algorithm can not only efficiently detect cuts and gradual transitions, but also is robust to motion.

Key words: B1ock motion vector estimation, Motion compensation, Self-adaptive dual thresholds, Shot segmentation