基于DC系数和运动矢量的快速场景分割算法

场景分割技术是动态视频分析和基于内容的视频检索的基础，以检测出来的场景作为基本单元，可以进一步对视频内容进行分析和建立索引．本文旨在提出一种基于MPEG压缩视频流的场景分割算法，利用MPEG数据流中已有的DCT DC系数和运动矢量，来检测场景的变换，从而实现场景分割，针对实际视频流中场景突变和渐变两类变换．本文提出两种方法分别处理不同情况，对于突变检测，该算法可以定位到帧，由于该算法进行最小程度的解码，降低了计算复杂度，因而大大提高了检测速度．

Fast Algorithm of Scene Change Detection Based on DC Coefficients and Motion Vectors

Scene change detection is one of fundamental processes in video analysis and content based video retrieval. Based on the basic unit of scene, the video content can be further parsed and indexed. This paper presents an algorithm on MPEG compressed streams to detect scene changes with DCT DC coefficients and motion vectors. Considering abrupt scene changes and gradual scene changes in real video sequences, two methods are developed to deal with different situations. For abrupt scene changes, the algorithm can locate the right frame. By performing minimal decoding on compressed bitstream, computational complexity is reduced, and the processing speed can be dramatically improved.