Research On Shot Boundary Detection Algorithms In Video Retrieval

Shot boundary detection technology is one of the key technologies of a content-basedvideo retrieval system . This paper presents a practical shot boundary detection algorithm.The proposed algorithm uses histogram to describe the content of the image. It calculatesa luminance histogram for each frame of the video stream. Based on the histograms, itcalculate a value to measure the differences of two consecutive frames, called"histogram-differences". And it defines a measure of the brightness level of a target image,called "average brightness." Thus, the sequence of frames corresponds to a sequence of"histogram difference", and a sequence of "average brightness".Based on the sequence of "histogram differences" , the proposed algorithm uses slidingwindow method to detect the hard cuts of the video. For a frame sequence without mutation,the "average brightness"-curve reflects the gradual transitons of the shots. The frames of asame shot correspond the flat part of the curve, and the frames of a gradual transitioncorrespond to the steep part of the curve. Thus, the proposed algorithm detects gradualchanges by judging the trend of the "average brightness"-curve.Brightness changes often cause false detection. We also present solutions for thisproblem. On the one hand, the proposed algorithm uses "average brightness" infomation toeliminate the impact of the flash of the images. On the other hand, based on the histogram, itcombines edge detection method to eliminate the impact of sudden changes in brightness.Therefore, the proposed algorithm includes the following two processes: for a video,firstly eliminate the impact of brightness changes, then detect hard cuts and gradualtransitions. The results of tests on the experimental data show that the algorithm can detectmost of the hard cuts and gradual transitions.