基于EM聚类的H.264压缩域视频对象实时分割算法

从压缩域直接分割视频对象比传统的像素域分割具有快速高效的特点,目前已有不少从MPEG域分割的方法,但从H.264压缩域分割的甚少。为此提出一种基于H.264域的实时分割运动对象方法,该算法先对当前视频帧进行全局运动估计和补偿,然后对4×4的运动矢量场进行分类处理,最后对非零运动矢量使用改进的EM聚类分割算法。本文算法对多个视频序列进行了实验,结果表明,该算法针对静止背景和运动背景的视频序列都能达到较精确的实时分割。

Real-time Video Object Segmentation in the Compressed Domain Based on EM Clustering

Moving object segmentation in the compressed domain is fast and efficient compared to former segmentation in pixel domain. Recently some moving object segmentation algorithm in MPEG compressed domain have been reported. However, very few approaches have been proposed in the H. 264 compressed domain. In this paper, a real-time video segmentation method in H.264 compressed domain is proposed. Firstly, the current frame is globally compensated in motion after the globalmotion estimation. Then the 4×4 motion vectors are processed bymedian filter and classification. Thirdly the non-zeromotion vectors are applied by improved EM clustering segmentation algorithm. The performance of the proposed approach is illustrated by simulation carried on several test sequences. Experiment results show that the proposed algorithm can provide satisfactory real-time segmentation results both on static background and moving background video.