Research On Video Shot Boundary Detection Algorithms

With the rapid development of Multimedia and network technology and increasingly wide applications of video, the expansion of the amount of video information has become a prominent problem. How to find interest content to users fast and effectively from mass of the video has become a serious problem. In this circumstance, the content-based video retrieval becomes one of the hot topics, while the shot boundary detection is basis of the content-based video retrieval. Shot boundary detection results have a direct impact on performance of content-based video retrieval.Content-based video retrieval theory and shot boundary detection technology are deeply discussed in this paper. Existing shot boundary detection technology and the problems are analyzed.The abrupt shot boundary detection is studied. Through the analysis of the existing shot boundary detection algorithm, histogram-based and based-pixel method can complement each other's advantages. In the process of the analysis for the characteristics of color space, each component of YUV color space has separable feature, contributing to reduce computation cost. And in the process of analysis on video we can find that inserting advertising or subtitles around the video, light and flash will result in adverse effects in the shot boundary detection. In order to improve the detection recall and precision, an abrupt shot boundary detection algorithm is implemented in the paper, and achieves good results. In the analysis of shot detection method we find that only detecting an type of shot boundary cannot meet the demand of shot detection. In addition, the existing methods are mostly for flash frame, and have not ruled out the impact of flashes sequence on detecting. According to the characteristics of flash sequence, we improve the algorithm based on abrupt shot detection, and eventually make the algorithm detect not only abrupt but also gradual shots in a test, while exclude the impact of flash sequence on shot boundary detection. Different types of video are experimented and confirm that this algorithm has better results.Through analyzing threshold-based shot detection methods, we can find the threshold selected directly affects the accuracy of detection results. The cluster-based approach can avoid the threshold selecting and does not need to learn. So the method can overcome the threshold selecting and versatility result from based-threshold and self-learning method. In addition, camera motion will bring interference on shot detection. In order to further improve the detection performance, combined fuzzy clustering with characteristics of camera motion, shot detection base on fuzzy clustering is reported in the paper. The algorithm needs not select threshold and reduces the impact of flash, subtitles and advertising on shot detection effectively. However, due to the complexity of the video itself, there are still missed and false detections in this algorithm and research is further needed.In the foreseeable future, the contents- based video retrieval will get more extensive application, and finally substitutes for existing text-based search mode.