Abstract: Shot boundary detection is the fundamental and the first step towards video processing and retrieval.All I-frames in compressed video are partly decompressed and features of their DC images are extracted.Shot boundary is generally detected by difference of I frames DC images’ character and then precisely located by macroblocks movement information of B or P frames in compressed video.Experimental results indicate that the algorithm presented in this paper is effective and precise and it can compensate the disadvantage of shot boundary detection in compressed domain.

Key words: shot boundary detection, cuts, gradual transition, Continuous Difference of DC-pictures（CDDC）, peak area

摘要： 镜头边界检测是视频处理与检索中最重要的第一步。对视频流直接在压缩域中部分解压I帧，提取其DC图像的特征，利用连续I帧的DC图像的特征差异粗略定位镜头的边界，再根据压缩域中P、B帧的各种类型的宏块数量变化精确定位镜头边界。实验结果表明，本算法能高效而精确地对突变镜头和渐变镜头进行检测，弥补了压缩域中镜头定位精确性不高的缺点。

关键词: 镜头边界检测, 突变, 渐变, 连续DC图像差值, 峰区