一种基于MPEG压缩域的视频对象提取方法  

MPEG Compressed Domain Video Object Extraction

作　　者：钟善君[1] 赖玉霞[1] 刘建平[1] 

机构地区：[1]浙江理工大学信息电子学院,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2006年第4期450-455,共6页Journal of Zhejiang Sci-Tech University(Natural Sciences)

摘　　要：为了加快视频对象的提取速度,提出了一种直接从MPEG压缩视频流中准确地提取视频对象的方法。首先从码流中提取运动矢量及DC系数,通过运动补偿建立DC Image;同时对运动矢量在时域上进行平均,在空域上进行带权矢量中值滤波处理。然后再定义了与运动矢量和DC系数都紧密相关的、以8×8块为基本单位的向量;之后对向量进行基于密度和层次的聚类算法得到分层的视频对象树,再通过计算最佳对象数目并提取视频对象。The authors propose a novel method to extract video object from MPEG compressed field to save time. First, motion vectors and DC. DC images are reconstructed by motion compensation technique. Motion vectors are dealt with temporal average filter and a spatially weighted vector median filter. Second, a vector is defined based on 8 by 8 pixel size, which is correlated to the DC and motion vectors. Finally, density based hierarchical clustering algorithm is applied to the new vectors and a hierarchical object partition tree is generated. The optimum number of objects is calculated.

关 键 词：视频对象 密度与层次聚类 运动矢量 MPEG 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]