一种基于过滤的快速视频镜头检测算法

A Fast Video Shot Detection Algorithm Based on Filter

视频镜头检测是基于内容视频检索的第一步，针对视频中的叠化和淡入淡出，提出一种基于三步策略的渐变镜头边界检测算法。提取视频帧的亮度和方差作为特征，通过有限状态机实现初始渐变检测，并计算视频帧的颜色、共生矩阵、运动特征来进行三步筛选，保证检测的准确性。对大量TRECVID视频进行实验并对比相关研究成果，结果表明该算法对渐变具有良好的检测性能，对运动及闪光具有较好的鲁棒性，能满足一些实时系统的需要。

Video shot detection is the first step of content based video retrieval, with regard to dissolve and fade in--fade out of video, This paper proposes the video shot detection algorithm based on three steps filter strategy. Firstly, a finite state machine is used to search gradual transition candidates in intensity variance continuity signal, and verify those to get real gradual transitions, next, extract color, texture and motion feature of video frame to guarantee the precision of gradual shot detection. Experimental results based on a large collection of videos from TRECVID demonstrate good performance of the algorithm and its robustness against disturbances caused by object motion or flashlight, very satisfy with real time system＇ s requirement.