An efficient detection algorithm based on anisotropic diffusion for low-contrast defect
An efficient detection algorithm based on anisotropic diffusion for low-contrast defect
复制标题
DOI：
10.1007/s00170-017-1156-6
复制
发表时间：
2018-02
期刊：
影响因子：
--
通讯作者：
Chin-Sheng Chen;Chi-Min Weng;Chien-Chuan Tseng
中科院分区：
文献类型：
--
作者：
Chin-Sheng Chen;Chi-Min Weng;Chien-Chuan Tseng
本文提出了一种基于各向异性扩散模型的低对比度表面图像缺陷检测算法，特别是针对增透（AR）玻璃。该算法包括两个重要步骤：（1）改进的各向异性扩散模型和（2）形态方向性滤波。为了快速有效地增强低对比度缺陷，提出了一种基于自适应边缘阈值（标准分数）的改进扩散模型。它充当自适应增强过程。具有低梯度和高Z分数或高梯度和低Z分数的像素将生成高扩散系数。它增强了疑似缺陷边缘的灰度级，并保留了疑似缺陷内部区域的原始灰度级，以及对噪声背景进行轻微的平滑处理。一个简单而有效的方法可以很容易地分割的缺陷，其次是有效的形态方向性滤波器，它消除了噪声的阈值图像。在这项研究中，所提出的算法进行评估的一组23个低对比度的表面图像的AR玻璃。实验结果表明，该算法优于其他四种竞争算法的上级。缺陷检测结果表明，该算法可以分割出完整的缺陷。此外，与Perona和Malik模型、C-T模型1、C-T模型2和直方图统计模型相比，计算优势分别约为4.48、2.26、19.53和26.63倍。实验结果表明，该算法不仅检测结果可靠，而且检测效率提高了2倍以上。
In this paper, we propose an efficient algorithm based on the anisotropic diffusion model to detect defect in a low-contrast surface image, especially aimed at anti-reflective (AR) glass. The proposed algorithm has two important procedures: (1) a modified anisotropic diffusion model and (2) morphological directivity filter. The modified diffusion model based on an adaptive edge threshold, the standard score (Z-score), is proposed to quickly and efficiently enhance the low-contrast defects. It acts as the adaptive enhancement process. The pixels with both the low gradient and the highZ-score or the high gradient and the lowZ-score will generate a high diffusion coefficient. It enhances the gray levels of suspected defective edges and also preserves the original gray levels of an internal area of suspected defects, as well as generates the slight smoothing process for the noisy background. A simple and efficient method can easily segment the defects, followed by the effective morphological directivity filter, which removes noise from the thresholding image. The proposed algorithm is evaluated by a set of 23 low-contrast surface images of AR glass in this study. The experimental results show that the proposed algorithm is superior to the four competitive approaches. The defect detection results demonstrate that the proposed algorithm can segment the complete defects. In addition, the computational advantage compared to the Perona and Malik model, C-T model 1, C-T model 2, and histogram statistics model are about 4.48, 2.26, 19.53, and 26.63 times, respectively. It can be concluded that the proposed algorithm not only provides reliable inspection results but also improves the inspection efficiency over 2 times.