一种改进的大津法在机器视觉中的应用  

Improvement on Canny operator by algorithm of self-adaptive determining double-threshold

作　　者：李牧[1] 闫继宏[1] 朱延河[1] 赵杰[1] 

机构地区：[1]哈尔滨工业大学机器人研究所,哈尔滨150080

出　　处：《吉林大学学报（工学版）》2008年第4期913-918,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：长江学者和创新团队发展计划项目(IRT0423);'863'国家高技术研究发展计划项目(2006AA04Z245)

摘　　要：应用Canny算子对梯度幅值图像进行模非极大值抑制后,需先确定高低阈值后才能进行边缘提取,而且不同的图像应当采用的阈值也不相同。目前双阈值的大小主要凭借个人经验确定,不具有科学性,从而限制了Canny算子在实际中的应用。为了对Canny算子的这一缺陷加以改进,提出了基于梯度幅度直方图和类内方差最小化原理自适应地求取双阈值的算法对大津法作出改进,理论推导和实验结果证明了该算法的正确性和有效性。Traditional Canny-operator has a satisfied chanracter in edge detection. But if gradient image is processed with module non-maximum suppression, double-thresholds should be evaluated to detect edges when Canny algorithm is adopted. For different images, the thresholds should not be identical. Currently the double-thresholds are normally obtained through personal experience, which restricts the use of Canny operator in practice. To solve this problem, an algorithm which can adaptively determine the double thresholds based on gradient histogram and minimum interclass variance and improve Canny operator was proposed. Theory and experiments show that the proposed algorithm is effective and correct.

关 键 词：信息处理技术 自适应确定双阈值 类内方差最小化 CANNY算子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]