Method for Video Shot Detection and Separation

Shot boundary detection is main step in video management systems like browsing and indexing. In this paper, we shortly describe an earlier proposed shot detection algorithm based on the structural properties of video frames. Two mathematical models for decision making, i.e. for similarity threshold determination are proposed and compared. The first model allows determination of threshold in case of using mean-square deviation for frames similarity and when the image pixel is assumed to be normal distributed random variable. The second model is based on using the structural similarity measure based on Weibull model for image gradient magnitude distribution. Global and adaptive approaches are considered for frames similarity threshold determination. Results of experiments to detect real video shots are given. It is shown that adaptive threshold determination method generally gives more acceptable results than global threshold determination method. At the same time the 2 W based threshold determination approach gives more accurate results than PSNR based approach, therewith corresponds to HVS perception.