基于相似性定位和超像素分割的织物疵点检测  

Fabric defect detection based on similarity location and superpixel segmentation

作　　者：朱磊[1] 任梦凡 潘杨[1] 李博涛 ZHU Lei;REN Mengfan;PAN Yang;LI Botao(School of Electronics and Information, Xi′an Polytechnic University, Xi′an, Shaanxi 710048, China)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048

出　　处：《纺织学报》2020年第10期58-66,共9页Journal of Textile Research

基　　金：国家自然科学基金项目(61971339);陕西省重点研发计划项目(2019GY-113);西安市科技局创新引导计划项目(201805030YD8C G14(6))。

摘　　要：为解决周期性纹理织物图像的疵点检测及其轮廓精确分割问题,提出一种基于相似性定位和超像素分割的织物疵点检测方法。将待检测图像进行中值滤波和对数增强,并利用FT算法估计增强图像的显著图实现待检测图像的预处理;将基于归一化局部均值差分的灰度相似性检测参量和结构相似性检测参量结合,构建可测量更多类型周期性纹理织物图像的相似性度量函数,通过阈值化增强图像分块的相似性测量值实现疵点在显著图中的粗定位;最后对显著图粗定位图像分块进行超像素细分割及其二值化处理,并借助连通域分析剔除孤立点,获得完整的疵点轮廓。结果表明,本方法与常规3种方法相比,对周期性纹理织物图像的疵点检测准确率更高,且提取出的疵点轮廓更精确。Aiming at the problem in defect detection and accurate contour segmentation of periodic texture fabric image,a method of fabric defect detection was proposed based on similarity location and superpixel segmentation techniques.The median filter and logarithm enhancement were applied for the detected image,and the saliency image of the enhancement image was estimated by frequency-tuned algorithm to facilitate the preprocessing of the detected image.Combining gray similarity detection parameters based on the normalized local mean difference with structural similarity detection parameters,a similarity metric function capable of measuring more types of periodic texture fabric images was constructed.The rough localization of defects was identified by thresholding the similarity measurement value of the enhancement image blocks.Finally,superpixel fine segmentation and binarization were performed on the rough localization image blocks,and the outliers were eliminated via connected domain analysis to obtain a complete defect contour.The experimental results show that,compared with the three conrentional methods,the proposed method has a higher accuracy in detecting the defects in the periodic texture fabric image,and the extracted defect contour is more accurate.

关 键 词：织物疵点检测 相似性定位 超像素分割 相似性度量函数 归一化局部均值差分 

分 类 号：TS101.9[轻工技术与工程—纺织工程] TP391.4[轻工技术与工程—纺织科学与工程]