一种改进的基于互信息量的镜头边界检测算法

镜头边界检测是基于内容视频检索的基础环节。由于视频类型与内容众多,目前镜头边界检测中存在阈值选取困难、查全率和查准率不高等问题。针对以上问题提出一种改进的基于互信息量的镜头边界检测算法,在字幕检测定位算法有效定位字幕区域的基础上,通过比较非字幕、非四角区域HSV空间直方图求取的相邻帧间互信息量的差异程度,实现镜头边界检测。实验表明,与当前应用最广泛且比较有效的双阈值算法相比,突变镜头检测综合性能平均提高12.4%,渐变镜头检测综合性能平均提高8.2%,通过自适应阈值的选取,有效解决了阈值依赖人工经验选取的问题;与当前已提出的基于互信息量的镜头边界检测算法相比,该算法降低了计算复杂度、几乎能检测所有的淡入淡出镜头边界,并使得镜头边界检测具有较高的查全率与查准率。

An Improved Shot Boundary Detection Algorithm Based on Mutual Information

Shot boundary detection is the basis of content-based video retrieval. At present, many types of video and numerous video content in the shot boundary detection cause the problem that the recall rate and precision rate are not high . In the light of this, an improved shot boundary detection algorithm based on mutual information is proposed, In the basis of the effective position of the caption area detected by the algorithm of video caption detection and localization,The algorithm can realize shot boundary detection by calculating the difference of mutual information between frames which are in the non-caption and non-triangular area about non-uniform block histogram in HSV space. Compared with the most widely used and more effective dual threshold algorithm,This experiment show that ,and the average performance of the algorithm is improved by 12.4%, the average performance of gradual shot detection is improved by 8.2%. Adaptive threshold selection, solve the problem of threshold dependent artificial experience selection effectively;Compared with the shot boundary detection algorithm based on mutual information,This experiment show thatthe algorithm can detect almost all fade in and out of the lens boundary,and bring higher recall rate and precision rate to the shot boundary detection.