摘要
由于线阵相机拍摄的图像光照不均、对比度低,使得钢轨表面离散缺陷检测成为机器视觉检测的难点,为此提出局部非线性对比度增强法和改进最大熵阈值分割法对钢轨图像进行预处理.该算法将局部区域内相对较低的灰度级映射到更低的范围,相对较高的灰度级映射到更高的范围,实现对比度拉伸;通过分析图像的目标熵、背景熵、灰度概率分布曲线,使用图像目标熵最大、目标概率较小的改进最大熵阈值分割法对图像进行分割,得到包含噪声相对较少的图像.实验结果表明,文中提出的非线性对比度图像增强算法简单、快速、有效,而且增强效果与光照无关;与原始的最大熵、目标最大熵、OSTU阈值分割法相比,改进的最大熵分割阈值较小,分割后的图像包含的噪声少;改进的预处理算法对测试图像的漏检率和误检率分别是6.2%和7.3%.
It is a challenge to detect discrete defects in a vision system because of illumination inequality and low contrast of rail images which obtained by linear array camera. This paper presents a local non- linear contrast （LNC） enhancement and an improved maximum entropy （IME） threshold segmentation to preprocess images. LNC enhances the contrast by mapping relatively low gray level to lower gray level, high gray level to higher gray level. By analyzing curves of objects entropy, background entropy and gray- level probability distribution of image, we proposed IME algorithm to segment the image which selects a threshold that maximizes the object entropy and meanwhile keeps the object proportion in a low level, therefore, the pre-processed images contain less noise. The experimental results demonstrate that the LNC algorithm is easy to implement and enhances the image fast and effectively. What＇s more, it is illumination independent. IME segments image with smaller threshold and less noise compared with maximum entropy, object entropy and OSTU threshold segmentation methods. The undetected rate and false detection rate of improved preprocessing algorithms for test images is 6.2% and 7.3%, respectively.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2014年第5期800-805,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(51065021)
关键词
机器视觉
钢轨表面缺陷
图像增强
最大熵
machine vision
rail surface defect
image enhance
maximum entropy
作者简介
袁小翠（1988-），女，博士研究生，主要研究方向为光机电一体化技术；
吴禄慎（1953），男，教授，博士生导师，论文通讯作者，主要研究方向为逆向工程及光机电一体化、虚拟现实技术；
陈华伟（1977-），男，讲师，博士。主要研究方向为图像处理与逆向工程．
