采用多尺度方向微分比率的角点检测算法  

A Corner Detection Algorithm Based on Multi-Scale Directional Differential Ratio

作　　者：王富平[1] 水鹏朗[1] 

机构地区：[1]西安电子科技大学雷达信号处理国家重点实验室,西安710071

出　　处：《西安交通大学学报》2016年第4期68-75,共8页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(61271295)

摘　　要：针对相邻图像结构间的相互干扰会导致角点错检率升高的问题,提出了一种基于多尺度方向微分比率(MDDR)的角点检测算法。该算法首先利用Canny边缘检测算法提取原始灰度图像的边缘轮廓;对于每一个轮廓像素,分别利用各向同性导数和各向异性高斯方向导数滤波器提取主方向及其正交方向上的方向强度微分信息;然后将3个尺度下微分比率融合为MDDR测度;最后进行阈值和非极大值抑制处理获得最终角点。不同于传统角点检测算法中只使用单一滤波器,MDDR算法中使用2种不同类型的滤波器,能在精确提取角点附近不同方向微分信息的同时避免邻近图像结构间的相互干扰,从而提高了角点定位准确性并增强了测度的噪声鲁棒性。实验结果表明,MDDR算法的平均检测准确率比点到弦距离累积算法提高了27.1%,并且平均错检率比残余面积算法和Gabor算法分别低28.4%和32.4%。A new corner detection algorithm using multi-scale directional differential ratio（MDDR）of images is proposed to solve the problem that the interference between adjacent image structures may lead to the increase of the false detection ratio.The algorithm firstly extracts the image edge contour of an original grayscale image using the Canny edge detection algorithm.The differential information along the principal direction and its perpendicular direction of each contour pixel is extracted by using the isotropic Gaussian directional derivative filter and the anisotropic Gaussian directional derivative filter,respectively.Then,the differential ratios at three scales are fused into the MDDR corner measure.Final corners are then obtained through thresholding the fused measure followed by the non-maximum suppression. The MDDR algorithm is different from traditional corner detection algorithms that use a single filter,it uses two different kinds of filters to precisely extract the differential information along different directions around the corners and to avoid the interference between adjacent image structures.Thus,the corner localization accuracy is improved,and the noise-robustness of corner measures is enhanced.Experimental results and a comparison with the chord-to-point distance accumulationalgorithm show that the average detection accuracy of the proposed algorithm increases by27.1%,and the average false detection ratio of the proposed algorithm is 28.4% and 32.4%lower than those of the residual area and Gabor detection algorithms,respectively.

关 键 词：边缘检测 多尺度 方向微分比率 角点检测 

分 类 号：TN911.7[电子电信—通信与信息系统]