摘要： 提出一种基于h.264压缩域视频流中运动对象的分割方法。对提取的运动矢量场进行噪声处理，在空间检测和校正、中值滤波等处理后，得到可靠的运动矢量场。为进一步增强可靠的运动向量，利用后向投影迭代技术，采用均值聚类算法获得视频对象分割。仿真实验证明，该方法可以得到良好的分割质量，且码流无须解码，实时性良好。

关键词: 压缩域, 运动对象分割, 运动矢量累积

Abstract: This paper propses a video moving object segmentation algorithm based on h.264 compressed domain. Motion vector field extracted from h.264 compressed stream is processed by noise processing and median filter. An iteratively backward projection scheme is proposed to obtain an accumulated motion vector filed. Moving object region is extracted in turn based on k-means cluster. Experimental results show moving object is segmented well without decoding video compressed stream.

Key words: compressed domain, moving object segmentation, motion vector accumulation

中图分类号: 

• TP301.6