基于CycleGan的动车零件图像高光消除方法
A Highlight Elimination Method for Train Parts Image Based on CycleGan
沈钧贤 1梅劲松 1王干 2陈苏扬3
作者信息
- 1. 南京航空航天大学自动化学院,江苏南京 211106
- 2. 南京拓控信息科技股份有限公司,江苏南京 210004
- 3. 南京地铁运营有限责任公司,江苏南京 210000
- 折叠
摘要
应用光学无接触方法检测动车零件损伤时,采集图像的照射光易在零件上造成镜面反射,使得采集图像上存在高光,不利于对损伤部位的辨识.提出一种基于循环对抗生成网络(CycleGan)和光照补偿算法结合的方法来修复动车轮部图像的高光部分.首先通过生成手段初步修复高光区域,然后改进Ret-inex光照补偿算法,实现对生成图像的进一步增强,达到高光消除的目的.实验结果表明,所提方法大幅改善了高光区域的像素灰度,使得该部分像素灰度接近图像整体像素灰度,同时强化了暗部细节和图像信息,提高了损伤部分的辨识度.
Abstract
When using optical non-contact methods to detect damage to motor vehicle parts,the illu-mination light of the collected images can easily cause specular reflection on the parts,resulting in high light on the collected images,which is not conducive to the identification of damaged parts.This paper pro-poses a method based on a combination of CycleGan and illumination compensation algorithm(Retinex)to repair the highlight part of the moving wheel image.First,the highlight area is initially repaired through generation means,then the Retinex illumination compensation algorithm is improved to further enhance the generated image.Experimental results show that the method used greatly improves the pixel grayscale of the highlight area,making the pixel grayscale of this part close to the overall pixel grayscale of the im-age,which achieve the goal of highlight removal.Meanwhile,it strengthens the details and image informa-tion of the dark parts,and improves the recognition of the damaged parts.
关键词
高光消除/图像处理/光照补偿/循环对抗生成网络Key words
highlight removal/image processing/light compensation/CycleGan引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据