目前大部分压缩域视频对象的分割方法主要面向MPEG系列视频标准，且算法建模复杂。为了解决这一问题，现提出了一种新的基于H.264/AVC的压缩域时空联合运动对象分割（TSMOS）算法。该方法主要利用压缩码流中的DCT系数和运动矢量信息进行对象分割，并首先利用相邻帧DCT系数之差提取运动对象轮廓，同时通过对轮廓进行形态学和抗噪声处理来得到粗糙的运动对象帧差掩码；然后采用运动向量归一化、噪声向量滤除、权值扩展向量中值(WEVM)滤波及前帧分割结果后向投影技术来得到对象的运动掩码；最后通过引入有效机制合并帧差掩码和运动掩码来分割运动对象。实验证明，该算法可取得较好的分割效果。

Most of the current moving object segmentation algorithms in compressed domain were focused on the MPEGx video standards which were complexity intensive due to the use of complex mathematical models. In order to resolve these problems, a new temporal-spatio method for moving object segmentation (TSMOS) in H.264/AVC compressed domain was proposed. DCT coefficients and motion vectors (MV) in bits stream were mainly used in this algorithm. Firstly, the differential mask of moving object was abstracted using the DCT coefficients difference (QDCD) between adjacent frames and followed by morphological processing. Then the methods of normalizing MV, weighted extended vector median (WEVM) filter on the MV fields and the existent segmentation results projection were utilized to obtain the moving mask of object. Finally, accurate moving objects segmentation results were achieved by combining the differential mask and the moving mask. The experiments on H.264/AVC test sequences demonstrated the validity of the proposed methods.