Research On Content-Based Fast Video Shot Boundary Detection

The development of digital information technology has brought moreand more content-rich and diverse videos into people's lives.How to quickly and accurately retrieve the video content of interest from massive video libraries becomes the key to video management and dissemination.Currently,content-based video retrieval methods are mainly achieved through the following steps:feature extraction,shot boundary detection,and key frame extraction.Among them,the shot boundary detection is the basic link,and the performance of the algorithm determines the quality of video retrieval to a great extent.This paper proposes an improved shot boundary detection algorithm based on mutual information for the problem of low recall and precision in the shot boundary detection algorithm.The algorithm chooses mutual information as the feature value,locates the caption region by the caption detection and localization algorithm,divides the video frame by screening the region,and further filters out the four corner regions of the video frame,and extracts non-subtitles,non-four corners.The difference value of the mutual information between adjacent video frames obtained from the regional HSV space histogram is used as the video interframe difference measurement value;afterwards,the abrupt shot boundary detection algorithm is set to effectively filter out the flash frame,the specific type of gradient start point and the gradual change.In the course of the special intervention frames such as false catastrophic points,the sharpness and precision of the mutation shot boundary detection algorithm are greatly improved;at the same time,the gradient frame boundary detection is performed on video frames,and different settings are made.Fade in,fade out the lens,and incorporate shot boundary detection algorithms to improve the recall and precision of the gradient shot boundary detection algorithm.Experiments show that,compared with the most widely used and effective dual-threshold algorithm,the overall performance of the mutation lens detection F₁ value increased by an average of 12.4%,and the average performance of the gradient lens detection performance F₁ value increased by 8.2%.Through the selection of adaptive thresholds,the problem that the threshold depends on the selection of artificial experience is solved.Compared with the current performance better lens boundary detection algorithm based on mutual information,the algorithm has low computational complexity and can detect all the edges of the fade-in and fade-out lens.The overall performance F₁ value of the mutant lens detection increased by an average of 5.8%,and the overall performance of the gradient lens detection F₁value increased by an average of 3.9%.