融合全局显著性信息的轮廓编组模型  

Closed Contour Extraction by Perceptual Organization and Global Saliency

作　　者：邓强[1,2] 罗予频[1,2] 葛俊锋[3] 

机构地区：[1]清华大学自动化系,北京100084 [2]清华信息科学与技术国家实验室,北京100084 [3]华中科技大学控制科学与工程系,武汉430074

出　　处：《计算机辅助设计与图形学学报》2013年第8期1223-1229,共7页Journal of Computer-Aided Design & Computer Graphics

摘　　要：基于感知编组的轮廓提取算法容易受背景上边缘的影响,导致轮廓提取的准确率低,为此提出一种结合感知编组与全局显著性信息的轮廓提取算法.首先在Canny算子框架下增加显著性信息的约束,提取显著边缘,减少了背景上的边缘;然后在Ratio-contour算法的基础上提出了新的目标函数,使得文中算法能够收敛于显著性高的区域,得到的轮廓更准确地标识前景物体.实验结果表明,该算法有效地提高了轮廓提取的准确性,同时大幅减少了轮廓提取的计算时间.In this paper a contour extraction algorithm combing perceptual organization and global saliency is proposed to reduce redundant edges and improve the precision of the extracted contour. The novelty lies in two aspects： firstly, a salient edge detection algorithm incorporating the saliency information is proposed to reduce redundant edges in the background; Secondly, a saliency based cost function is introduced to make the algorithm converge at regions with high saliency, which could better mark the foreground object. Experiments show that the proposed algorithm could improve the accuracy of contour extraction, meanwhile, reduce the computation time efficiently.

关 键 词：感知编组 全局显著性 显著边缘 轮廓提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]