A System of Video Shot Detection Using Multi-stage Algorithm

Shot detection is the first and important stage of video analysis. In this paper, an effective shot change detection system using multi-stage algorithm is presented. First the isolated clips were founded in a given video sequence, and corresponding three dimensional wavelet transform representations of these image sequences were obtained. Then shot detection of multi-scale and controllable precision was carried out. At the lowest level, the cuts were detected through the histogram difference, while the gradual transitions were detected through the histogram accumulation difference. Based on the rough detection result, the matrixes of variant size´s regions in wavelet representation of different levels were carried on the singular value decomposition, and then the dissolves and fades were classified using the angle of the singular value decomposition. And the wipes were detected by Gaussian weighted Hausdorff distance, this method can not only detect wipe transitions but also determine the types of wipes effectively. Therefore the precision controllable detection results were obtained. Simulation results show that the system can produce accurate shot change detection, which is very useful for video retrieval.