基于邻域特性分析的小模数齿轮亚像素图像边缘检测

Sub-pixel image edge detection based on neighborhood characteristic analysis for small modulus gear

针对高分辨率小模数齿轮图像中边缘灰度梯度呈双峰或多峰分布的问题,提出一种基于邻域特性分析的小模数齿轮图像亚像素边缘检测算法。算法以邻域窗口的灰度均方差积表示边缘强度、灰度重心所在的方向表示灰度变化的方向;通过阈值化、非极大值抑制和边缘细化等获取初始边缘;在初始边缘的基础上将求取的灰度变化方向划分为八个区域,构建一维灰度矩模型解算亚像素边缘位置。为验证算法的有效性,分别以模拟图像与实际图像为对象进行亚像素边缘检测实验,根据齿轮边缘特性以最小二乘拟合法评价算法对齿轮图像的检测效果。实验结果表明,该算法具有曝光敏感度优、绝对定位精度高、重复性好的优点。

Aiming at the problem that in high-resolution small modulus gear image the edge gray gradients show double or multimodal peaks,a sub-pixel edge detection algorithm based on neighborhood characteristic analysis is proposed for small modulus gear image.The algorithm uses the product of the gray mean square deviation in neighborhood window to represent the edge intensity and uses the directions of gray centroid to represent the directions of the gray change.Then,the initial edge is obtained using thresholding processing,nonmaximum suppression and edge thinning.On the basis of the initial edge,the obtained gray change directions are divided into eight regions; a 1 D gray moment model is built to solve the sub-pixel edge position.In order to verify the effectiveness of the algorithm,taking the simulated image and actual image as targets,sub-pixel edge detection experiments were conducted.According to the gear edge characteristics,the least-square fitting method was used to evaluate the gear image detection effect of the algorithm.The experiment results indicate that the proposed algorithm has the advantages of better exposure sensitivity,high absolute positioning precision and good repeatability.