【摘要】 镜头边界的自动检测是实现基于内容的视频检索必不可少的第一步 ,目前大多数的场景转换检测方法都是基于非压缩视频的 ,而越来越多的视频数据却以压缩形式存在。本文提出了一种新的针对MPEG压缩视频的场景转换检测算法 ,它利用DC序列和运动向量计算像素差、直方图差、统计差和具有“真实”运动向量的宏块所占的比例 ,然后用模糊逻辑对上述参量加以综合 ,隶属度用自适应的方法确定。实验表明这种镜头检测算法具有较高的检出率和检测精度 ,并且能适应不同类型的视频流。更多还原

【Abstract】 The auto-detection of shot boundary is a necessary step for content-ba sed video retrieval Most scene change detection methods are based on non-compr e ssed video,but more and more video are stored in compressed mode We introduce a new scene change detection algorithm for MPEG compressed video The algorithm us e s DC sequence and motion vector to calculate pixel differences,histogram differe nces,statistical differences and the ratio of macroblocks with true motion vecto r Then a fuzzy logic system with adaptive member function is used to conclude t h e above metrics Experiments show that this algorithm can detect shot boundaries with high accuracy and has good adaptability to different types of video更多还原