光照不均匀下的工件图像亚像素边缘检测  

Subpixel edge detection of workpiece images under uneven illumination

作　　者：刘红涛 陆安江 彭熙舜 杨教 龙纪安 Liu Hongtao;Lu Anjiang;Peng Xishun;Yang Jiao;Long Ji'an(School of Big Data and Information Engineering,Guizhou University,Guizhou 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州550025

出　　处：《国外电子测量技术》2022年第10期1-7,共7页Foreign Electronic Measurement Technology

基　　金：国家自然科学基金-地区项目(61865002);黔科合平台人才[2018]5781号项目资助。

摘　　要：针对光照不均匀环境下传统边缘检测算子出现的边缘检测模糊、自适应差等问题,提出了一种基于Retinex理论和改进的高斯拟合亚像素边缘检测算法。首先采用Retinex理论对照度分量和反射分量同时进行处理;其次利用传统的Canny算子获取图像边缘的像素坐标点,改进梯度矩阵来选取拟合点后找到梯度矩阵的切线方向,之后利用法线上的像素点计算所得的梯度值进行高斯拟合获取亚像素点坐标;最后利用八邻域查找连接每个轮廓的亚像素点并输出工件边缘图像。实验结果表明,相较于Zernike矩算法和其他两种高斯拟合算法,算法在4组图像实验中的Pratt品质因数、均方误差、峰值信噪比、结构相似性指标的均值为0.727、11 980、7.64、0.093,各组数据均优于其他3种算法,具有一定的普适性。In view of the blurred edge detection and poor self-adaptation of the traditional edge detection operator in the uneven illumination environment, this paper proposes a subpixel edge detection algorithm based on Retinex theory and improved Gaussian fitting. Firstly, the Retinex theory is used to process the illuminance component and the reflection component at the same time. Secondly, the traditional Canny operator is used to obtain the pixel coordinate points of the image edge, and the gradient matrix is improved to select the fitting points to find the tangent direction of the gradient matrix. Then use the gradient value calculated by the pixels on the normal to perform Gaussian fitting to obtain sub-pixel coordinates. Finally, use eight neighborhoods to find the sub-pixel points connecting each contour and output the workpiece edge image. The experimental results show that, compared with the Zernike moment algorithm and the other two Gaussian fitting algorithms, the mean value of Pratt′s figure of merit, mean square error, peak signal-to-noise ratio and structural similarity index in the four groups of image experiments is 0.727, 11 980, 7.64, 0.093. Each group of data is better than the other three algorithms, and has certain universality.

关 键 词：RETINEX理论 梯度矩阵 高斯拟合 亚像素边缘 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]