摘要
针对前景与背景具有相似颜色时的运动对象分割问题,提出一种结合核密度估计和边缘信息的分割算法.在前景和背景建模阶段使用颜色信息的基础上,引入边缘信息来构造前景和背景的概率模型;然后在马尔可夫随机场框架下引入与概率模型有关的似然能量项,以及反映空域连续性和时域一致性的能量项,并利用图切割方法来获得可靠的运动对象分割结果.实验结果证明,对于前景与背景具有相似颜色的视频序列,该算法降低了对象分割误差,显著地提高了整个序列中对象分割的鲁棒性.
This paper proposes a novel moving object segmentation algorithm based on kernel density estimation and edge information to solve the color similarity problem between foreground and background. During the stage of foreground/background modeling, both color feature and edge feature are used to build two probability models. Under the framework of Markov random field （MRF）, three energy terms associated with the likelihood of foreground/background, spatial continuity and temporal consistency are introduced to construct a graph, and the graph cut method is exploited to reliably segment moving objects. Experimental results demonstrate that the proposed algorithm reduces the segmentation error when foreground and background show similar colors, and greatly enhances the segmentation robustness during the whole video sequence.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2009年第2期223-228,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60602012)
上海市教育发展基金会晨光计划项目(2007CG53)
上海市教育委员会科研创新项目(09YZ02)
上海大学优秀青年教师基金
关键词
运动对象分割
核密度估计
马尔可夫随机场
moving object segmentation
kernel density estimation
Markov random field （MRF）
