张映权,王琼华,李大海,张文涛

摘 要:SUSAN角点检测算法以抗噪声性能强,运算速度快而被广泛运用于特征点的提取。传统的SUSAN算法的灰度差阈值固定,不能有效去除伪角点,并且在大尺寸模板检测下耗时多。针对这些问题,从模板尺寸对检测结果的影响出发,讨论不同尺寸模板的检测效果,从而提出一种变换模板提取特征点的方法。采用一种自动选取阈值的方法实现了阈值的自动选取,使用能量分布法和像素投影法去除了伪角点。结果显示,该方法缩短了检测时间,并且提高了检测准确度。

关键词:特征提取;SUSAN算法;能量分布;像素投影

中图分类号:TP391文献标识码:A

文章编号:1004-373X(2009)20-042-03

Improved SUSAN Corner Detection Algorithm

ZHANG Yingquan,WANG Qionghua,LI Dahai,ZHANG Wentao

(School of Electronics and Information Engineering,Key Laboratory of Fundamental Synthetic Vision Graphics and Imagefor National Defense,Sichuan University,Chengdu,610065,China)

Abstract:SUSAN corner detection algorithm is widely used in feature extraction for its good performance in noise resistance and fast calculation.The traditional SUSAN algorithm has a fixed brightness difference threshold and can′t eliminate the fake corner well.The traditional algorithm is time-consuming when large-size mask is used.Aiming at those problems,the relationship between mask size and detection results is discussed,and an algorithm using alternate mask is proposed.A method that can select the threshold automatically is adopted.The energy distribution and pixel projection methods are used to eliminate the fake corners.The experimental results show that this improved algorithm reduces the detection time and improves the detection accuracy.

Keywords:feature extraction;SUSAN algorithm;energy distribution;pixel projection

0 引 言

在计算机视觉和图像处理中角点还没有明确的数学定义,存在多种数学描述方法。一般的角点定义为图像中灰度的剧烈变化点或是曲线曲率的极大值点[1,2]。角点包含了大量图像信息,但却减小了计算量[3,4],因而被广泛运用于图像识别、目标匹配、图像融合以及三维重建之中[5,6]。SUSAN (Smallest Univalue Segment Assimilating Nucleus)算法[7,8]是由Smith等人提出的一种底层图像处理方法。此算法简单,定位精度高,运算速度快,其不需要梯度运算的特点,使该算法有很好的抗噪声性能[9,10]。但SUSAN算法也存在一些不足,例如,当模板增大时,其运算量显著增加,同时它不能很好地去除伪角点。基于以上两点,这里提出一种变换模板的方法,解决在大模板下耗时的缺点;然后使用能量分布和像素投影的方法去除伪角点。

1 SUSAN算法及改进算法

1.1 传统SUSAN算法的基本原理