基于 MPEG 国际标准压缩视频流的镜头切分算法

Approach for shot segmentation using MPEG compressed data

镜头切分是实现对动态视频基于内容检索的第一步，以检测出来的镜头作为基本单元，可以进一步对视频内容进行分析和建立索引。从实用角度看，目前越来越多的动态视频资料都是以压缩形式存储和传输，所以，研究基于压缩视频流的算法更有实际意义。本文旨在提出一种基于ＭＰＥＧ国际标准压缩视频流的镜头自动切分算法，通过利用ＭＰＥＧ数据流中已有的信息，如离散余弦变换（ＤＣＴ）系数和运动向量，只进行最小程度的解码，来检测镜头间的边界，从而实现镜头切分。针对实际视频流中镜头切换方式的复杂性，本文提出了三个算法分别处理不同情况，并将这三个算法以树形分类器的方式组织在一起，形成一个系统。通过对十段不同类型的ＭＰＥＧ－Ⅰ压缩视频节目进行镜头切分实验，取得了９０％以上的正确率。

Segmenting video sequences into individual shots is one of the fundamental processes in content based video retrieval. We can further parse and index the video content based on the basic unit of shot. Up to now, more and more video materials are stored and transmitted in the compression form, so it is practical to study the shot segmentation algorithms based on compressed video data. This paper presents an integrated approach to detect the boundaries between shots by using the discret cosine transform (DCT) coefficients and motion vectors encoded in MPEG compressed data. Only minimal decoding is needed for the algorithm. Considering the complicated situation in real world video sequences, three algorithms are developed to deal with different situations, and we also present a tree like classifier to organize the three algorithms together to form a system. By testing ten video sequences for various types, we get over 90% correct percentage.