基于总变差的织物疵点分割方法  

Fabric defects segmentation using total variation

作　　者：刘国维 潘如如[1] 高卫东[1] 周建[1] LIU Guowei;PAN Ruru;GAO Weidong;ZHOU Jian(Key Laboratory of Eco-Textiles( Jiangnan University), Ministry of Education, Wuxi, Jiangsu 214122, China)

机构地区：[1]生态纺织教育部重点实验室(江南大学),江苏无锡214122

出　　处：《纺织学报》2021年第11期64-70,共7页Journal of Textile Research

基　　金：国家自然科学基金项目(61501209)。

摘　　要：为解决当前机织物疵点检测方法精度不足的问题,提出了基于总变差模型的织物疵点分割方法,并着重分割经纬向尺寸小且异常不显著的疵点。首先应用奇异值分解低秩重建的方法将织物纹理背景去除,获取疵点异常图;然后通过构建总变差模型对疵点异常图进行最优化求解处理,得到不同约束下的疵点增强图;最后通过常规分割算法实现疵点的准确分割。实验结果表明:经总变差模型处理后的疵点异常图,其疵点与背景的可分割性得到显著提升。通过讨论总变差模型的参数对分割结果的影响,进一步验证了基于总变差的织物疵点分割方法的有效性和稳定性。Aiming at the problem of insufficient accuracy of the current woven fabric defect detection methods,this paper proposes a fabric defect segmentation method based on the total variation model,focusing on solving unobvious defects along warp and weft directions.The singular value decomposition low-rank was used to reconstruct the textile texture image by removing the texture background of the fabric so as to obtain the defect abnormal map.Following that,by constructing a total variation model,the defect anomaly map was optimized to obtain the enhanced abnormal map under different constraints.Finally,the accurate segmentation of defects was achieved through adopting the conventional segmentation algorithms.The experimental results show that the separability of the defects and the background of the defect abnormal image processed by the proposed total variation model has been significantly improved.The influence of the parameters of the model on the segmentation results was discussed to further verify the effectiveness and stability of the method.

关 键 词：织物疵点 纹理低秩重建 总变差模型 疵点分割 图像识别 

分 类 号：TS101.91[轻工技术与工程—纺织工程]