基于镜头关键帧集的视频场景聚类的研究

A Study of Video Scenes Clustering Based on Shot Key Frame Sets

在数字视频的分析、浏览、检索中,镜头已难以满足现有的需要。场景是一组包含有内容相关的若干镜头的集合,在一定程度上满足了数字视频的分析、浏览、检索的需要。文章首先使用了X2直方图匹配的计算方法,结合直方图的两次判断法,进行突变和渐变镜头边界的检测;然后对镜头内非相邻帧间距离经过阈值判断提取关键帧集;文章提出了基于镜头关键帧计算两个关键帧集之间距离的最小值作为所计算镜头之间的距离的算法;最后运用镜头之间的距离进行镜头的聚类产生场景,给出了典型的实验结果,表明该算法对视频场景的聚类有较好的性能。

In digital video analyse,browse,retrieve and query,shot is incapable of meeting of needs.Scene is a cluster of a series of shots,it partially meets the demand of digital video analyse,browse,retrieve and query.An algorithm of video scenen clustering based on shot key frame set is proposed.The X 2 histogram match and twin histogram comparison is used for shot detection.A method of extracting shot frame set based on distance of non adjacent frames is presented,further more,compute the minimum of distance of key frame sets as distance of shots,eventually cluster scenes accord-ing to the distance of shots.The experiment results of cluster scenes in digital video with this algorithm indicate that this algorithm has satisfactory performance of correctness and compute speed.