基于熵能的H.264压缩域运动对象分割

提出了一种基于熵能选取自适应阈值的时空域运动对象分割方法。首先对H.264压缩码流中提取的原始运动矢量场进行连续多帧的累加来增强运动信息，并对累积运动矢量场进行相似性判断，初步获得运动块；然后提取压缩码流中4×4块残差编码位数，并基于熵能自动选取自适应局部阈值，获取运动区域的轮廓信息；最后结合运动块和轮廓信息按照一定的规则对边界进行校正。对多个视频序列进行了实验，结果表明，该算法能快速取得较好的分割结果。

Video object segmentation in H.264 compressed domain based on entropy energy

The paper presented a new temporal-spatial method for moving object segmentation in H.264 based on local self-adaptive threshold on entropy. The Motion Vector (MV) fields of several continuous frames were accumulated to enhance the motion information. Then the similarity measure was performed on the accumulated MV field to remove part of noise. And then residual size of 4×4 block was extracted from the compressed bitstream, and each 4×4 block threshold was selected by local self-adaptive threshold on entropy. Finally according to some formula, the article ulteriorly refine the boundary of the motion block. The H.264 sequences test demonstrates the validity of the proposed method.