基于运动特征的H.264压缩域对象分割

Motion-characteristic based object segmentation in the H.264 compressed domain

基于从H.264压缩域提取的运动场,提出了一种新的运动对象分割方法。首先采用矢量中值滤波方法滤除运动场中的噪声矢量;再运用后向估计的方法重建预测运动场并进行运动场的累积;然后对存在背景运动的累积运动场进行全局运动补偿;最后基于幅度、散度和旋度3个运动特征,采用改进的统计区域合并方法将运动对象分割出来。实验结果表明,本文方法适用于背景静止或背景运动的H.264压缩视频,且分割质量较好。

A new moving object segmentation approach based on the motion field extracted from H. 264 compressed domain is proposed. The noisy motion vectors in the motion field are first removed by the vector median filtering. Then the predicted motion fields reconstructed by backward estimation are used to accumulate the motion field. After that ,global motion compensation is performed on the accumulated motion field with moving background. Finally the modified statistical region merging is exploited to segment the moving object based on magnitude, divergence and curl. Experimental results demonstrate that our approach is applied to the H. 264 video with static or moving background and the segmentation quality is good.