基于机器学习的边缘检测方法研究  

The research of edge detection based on machine learning

作　　者：赵彤洲[1,2] 王海晖[1,2] 徐迪迪[1,2] 

机构地区：[1]武汉工程大学智能机器人湖北省重点实验室,湖北武汉430073 [2]武汉工程大学计算机科学与工程学院,湖北武汉430073

出　　处：《湖北大学学报（自然科学版）》2011年第3期370-372,382,共4页Journal of Hubei University：Natural Science

基　　金：湖北省自然科学基金(2008CDB333)资助

摘　　要：传统的边缘检测方法具有一定的局限性,且自适应能力差,提出一种基于机器学习的边缘检测方法来解决上述问题.实验图像从伯克利图像数据库中选取,以Harr和梯度直方图(HoG)构成特征空间,将AdaBoost算法和决策树算法相结合进行分类器训练.实验结果表明,机器学习的边缘检测算法有更高的分类准确率.Because of the limitation and less adaptability on the traditional method of the edge detection, this paper proposed a method based on machine learning to solve such problems. The experiment images were selected from Berkeley Segmentation Dataset. The Harr and histogram of gradient etc effective features were selected to composie the feature space, and during the process of classifier training, combined the AdaBoost and decision tree algorithm to improve the classification accuracy. The experimental results indicated that machine learning algorithm for edge detection had the higher classification accuracy than the traditional one.

关 键 词：边缘检测 机器学习 特征提取 ADABOOST算法 决策树算法 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]