基于压缩域的视频关注对象提取  

An Approach to Video Object Extraction in the Compressed Domain

作　　者：潘琤雯[1] 张兆扬[1,2] 石旭利[1,2] 沈礼权[1,2] 

机构地区：[1]上海大学通信与信息工程学院,上海200072 [2]新型显示技术与应用教育部重点实验室,上海200072

出　　处：《中国图象图形学报》2009年第10期2145-2148,共4页Journal of Image and Graphics

基　　金：国家自然科学基金项目(60832003)

摘　　要：针对视频压缩域的对象分割问题,提出了一种视频对象精确提取方法。先将视频对象的颜色空间进行均值偏移和区域生长,得到分块的视频图像。同时对视频编码过程中所产生的运动矢量统计熵值,提取出视频中能引起人眼关注的运动对象。最后利用提出的对象轮廓精确分割方法,提取视频关注对象。实验表明该算法能精确和完整地分割出视频关注对象,并对视频中关注对象的切换、物体的非刚性形变有很好的鲁棒性。In this paper, we propose a new method to segment video object. Firstly, we smooth each frame with the meanshift and region-growing method. And then, the motion vectors which are released by video coding are used to extract a moving object. Finally, the object segmentation method proposed in this paper is used to extract the video attention object. Experimental results show that the proposed method can effectively separate the video object from the sequence and has the strong robustness in multi-targets tracking especially for switching targets and deformed objects.

关 键 词：均值偏移 信息熵 运动矢量 对象提取 

分 类 号：TN941.1[电子电信—信号与信息处理]