摘要
基于颜色恒常性理论的亮度感知Retinex模型,在频域采用多尺度高斯函数估计出输入图像的光照分量,对其进行伽马变换以增强光照信息,然后由输入图像与光照信息图像的对数运算获得仅包含物体本身特性的反射信息图像,实现对输入图像的校正。对校正后的图像依次进行频域带通滤波、基于全局的自适应阈值分割、形态学处理、亚像素精度的XLD轮廓提取等操作,根据提取结果来判断陶瓷基片表面有无缺陷。实际测试结果表明:该方法判断的准确度约为98.6%,检测时间约为0.21 s,满足工程需求。
Based on the brightness perception Retinex model of the color constancy theory, the multiscale Gaussian function was used in the frequency domain to estimate the illumination component of the input image. The gamma transform was performed to enhance the illumination information. The reflection information image that only contains the characteristics of the object itself was obtained by the logarithmic operation of the input image and the illumination information image to realize the correction of the input image. Frequency domain band-pass filtering, global adaptive threshold segmentation, morphological processing, sub-pixel precision XLD contour extraction and other operations on the corrected image were performed in sequence, and the ceramic substrate defect was determined by the extraction results. Actual testing shows that the accuracy of this method is about 98.6%, and the detection time is about 0.21 seconds, which meets engineering requirements.
作者
柴进平
罗敏
孙国浩
谢石磊
Chai Jinping;Luo Min;Sun Guohao;Xie Shilei(Hubei University of Automotive Technology,Shiyan 442002,China;Henan Nuomi Intelligent Technology Co.Ltd,Zhengzhou 450000,China)
出处
《湖北汽车工业学院学报》
2022年第3期56-60,共5页
Journal of Hubei University Of Automotive Technology
关键词
多尺度高斯函数
伽马变换
带通滤波
自适应阈值分割
multi-scale Gaussian function
Gamma transform
band-pass filtering
adaptive threshold segmentation
