基于改进型Retinex算法的轴承滚子瑕疵检测系统设计  

A Novel Bearing Roller Blemish Detection System Based on Improved Retinex Method

作　　者：刘超 刘畅[2,3,4] 余翔[1] 张玉成 易礼燕[1,2] 

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆 [2]移动计算与新型终端北京市重点实验室(中国科学院计算技术研究所),北京 [3]中国科学院计算技术研究所无线通信技术研究中心,北京 [4]中国科学院大学,北京

出　　处：《计算机科学与应用》2019年第2期239-249,共11页Computer Science and Application

摘　　要：基于机器视觉的工件瑕疵检测技术已经成为了重要的发展趋势,然而在实际的检测环境中,光照不均问题对检测结果带来了非常大的影响。因此,光照校正处理成为了机器视觉检测的关键步骤,但是现有的光照校正算法往往存在图像细节丢失的问题,直接影响了检测效果。针对这一问题,提出了一种改进型的Retinex光照校正算法,采用中值滤波代替了高斯环绕函数,从而实现了光照分量的估计,并使用信息熵函数完成了对原图光照均匀程度的判定。同时,基于所提算法搭建了一套轴承滚子瑕疵检测系统,并对所提算法的性能进行验证。实验结果表明,与现有的Retinex系列光照校正算法相比,该算法能够更好地保留工件的图像细节,从而实现更优的检测准确率。The workpiece blemish detection technology based on machine vision has become an important development trend. However, in the actual detection environment, Illumination unevenness has a very large impact on the detection results. Therefore, the illumination correction processing has become a key step in machine vision detection, but the existing illumination correction algorithms often have the problem of image detail loss, which directly affects the detection effect. Aiming at this problem, we proposes an improved Retinex illumination correction algorithm, which uses median filtering instead of Gaussian surround function to realize the estimation of illumination component, and uses the entropy function to complete the illumination uniformity of the original image determination. At the same time, based on the proposed algorithm, a set of bearing roller blemish detection system is built and the performance of the proposed algorithm is verified. The experimental results show that compared with the existing Retinex series illumination correction algorithm, the proposed algorithm can better preserve the image details of the workpiece, thus achieving better detection accuracy.

关 键 词：光照校正 中值滤波 机器视觉 缺陷检测 Retinex原理 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]