中文摘要：

视频数据大都是经过压缩域的形式存储和传输的，且直接在压缩域进行视频对象分割无需运动估计等复杂的计算，速度较快。本文提出了一种基于梯度模型的MPEG压缩域的运动对象分割算法。首先利用DCT（Ac[1]和Ac[8]）系数获得所有物体的边缘，然后综合在累积运动矢量基础上得到的边缘运动信息，从而获得感兴趣运动物体的边缘。仿真实验结果表明，它可以取得满意的分割质量。

英文摘要：

There are several advantages to perform video object segmentation in a compressed domain. The primary benefit is with much high processing speed. Moreover, the video data already exiting in MPEG1 and 2 formats in many databases would have to be decompressed to employ the pixel-domain techniques. In this paper, a moving objects segmentation algorithm based on gradient model in compressed domain is proposed. Firstly, DCT coefficients （AC[ 1 ] and AC[ 8 ] ） are utilized to form the gradient image. Then we synthesize the edge motion message obtained from the accumulated motion vectors. Finally, desirable moving objects segmentation results are acquired.