基于分布匹配的主动轮廓模型及其图像分割算法  

Active contour model based on distribution matching and its image segmentation algorithm

作　　者：陈宗海[1,2] 方伟[1] 陈会勇[1] 王智灵[1] 

机构地区：[1]中国科学技术大学自动化系多媒体计算与通信教育部一微软重点实验室,合肥230027 [2]中国科学院自动化研究所模式识别国家重点实验室,北京100080

出　　处：《吉林大学学报（工学版）》2008年第6期1441-1446,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：“863”国家高技术研究发展计划项目(2007AA04Z227);多媒体计算与通信教育部-微软重点实验室开放基金项目(05071806);模式识别国家重点实验室开放课题基金项目(2006-3)

摘　　要：针对运动目标跟踪的快速性和鲁棒性要求,提出了基于分布匹配的主动轮廓模型,该模型将模板和候选区域的分布匹配标准——相对熵(即Kullback-Leibler距离)作为主动轮廓的外部能量,使用目标和背景全图像的匹配,克服了跟踪过程中噪声的影响。同时根据匹配力构建帧内和帧间的运动向量,加快了序列图像分割速度。相比于现有的基于边缘、块匹配、区域匹配的主动轮廓模型,该模型具有更好的噪声鲁棒性及在跟踪过程中容忍目标旋转和较大位移的能力。实验验证了该模型的有效性。A new active contour model based on distribution matching is proposed, In which the speediness androbust demands for motion object tracking are taken into account. In this model the Kullback-Leibler divergence is regarded as the external energy of the active contour. The influence of noise in the obiect tracking process can be overcome by using the distribution matching of obiect and background. Also the motion vector established from the matching force can accelerate the sequential image segmentation. Compared with existing methods, such as edge based, block matching based, region matching based active contour models, the proposed model is more robust to noise, and more tolerant to object rotation and larger displacement in the tracking process. The efficiency of the proposed model is verified by experiments.

关 键 词：计算机应用 分布匹配 主动轮廓模型 序列图像分割 运动目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]