局部运动场景中运动对象的压缩域分割算法  

Compressed Domain MoVing Object Segmentation in Local Motion Scene

作　　者：刘龙[1] 韩崇昭[1] 刘洁瑜[1] 王占辉[1] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049

出　　处：《电子与信息学报》2005年第8期1249-1253,共5页Journal of Electronics & Information Technology

基　　金：国家973计划(2001CB309403)资助课题

摘　　要：压缩域视频运动对象的分割技术近年来逐渐被人们所重视,并且成为较实用的技术之一。该文针对目前压缩域中运动对象分割质量不高的问题,提出一种压缩域分割的改进算法。该算法首先从MPEG压缩码流中提取运动矢量,并对矢量进行运动累加和中值滤波处理,最后利用贝叶斯风险决策对非零运动矢量区域进行校正,最终得到较好的分割结果。该文还定义了压缩域分割质量的评价参数分割率和错分率,为评价压缩域的分割结果提供了客观评价标准。More and more attention has been paid to the study of moving object segmentation in MPEG compressed domain these years, which has already developed into a practical technology. In this paper, an algorithm by using motion vectors and DCT coefficients is put forward for segmenting moving object in MPEG compressed domain. As regards local motion scene, the algorithm produces a better result of segmentation than currently adopted ones. Firstly, the motion vectors are processed by accumulation and median filter; then the areas of non-zero motion vectors are rectified by the Bayesian decision to get high-quality segmentation.

关 键 词：MPEG 分割 压缩域 

分 类 号：TN919.8[电子电信—通信与信息系统]