应用单演小波分析的织物疵点检测  

Fabric defect detection using monogenic wavelet analysis

作　　者：厉征鑫[1] 周建[1] 潘如如[1] 刘建立[1] 高卫东[1] 

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》2016年第9期59-64,共6页Journal of Textile Research

基　　金：国家自然科学基金项目(61501209);国家自然科学青年基金项目(61203364);高等学校博士学科点专项科研基金项目(20120093130001);江苏高校优势学科建设工程资助项目(苏政办发[2014]37号)

摘　　要：为解决现有织物疵点检测算法对种类繁多的疵点形式尤其是对微弱纹理变化疵点的适应性较弱问题,提出以单演小波分析工具为基础的织物疵点检测算法。通过分数阶拉普拉斯算子与多重调和样条构建各向同性拉普拉斯小波后,对其进行Riesz变换生成Riesz-拉普拉斯小波,实现了织物图像的单演小波分析。对单演小波分析结果中的多分辨率方向与振幅子带,分别设计了最优响应判断标准以及最优响应子带分割方法。实验结果表明,所提出的检测算法能有效分割不同织物纹理中的多种类疵点,分割结果可反映疵点位置与轮廓,对342幅实验样本图像实现了97.37%的检出率,具有较好的自适应性与鲁棒性。In order to overcome the poor adaptability of existing fabric defect detection algorithms on numerous kinds of defects, especially minor texture changes, a fabric defect detection algorithm based on monogenic wavelet analysis was proposed. The monogenie wavelet analysis on fabric images works with the Riesz-Laplace wavelet, which is generated by performing Riesz transform to an isotropic Laplace wavelet constructed by combining a fractional Laplacian with a polyharmonie spline. For the mnltiresolusional orientation and amplitude subbands outputted by monogenic wavelet analysis, respective criteria for the best responses and segmentation method on the best response subbands were designed. Experimental results showed that the proposed detection algorithm could effectively segment various kinds of defects in different fabric textures, consequently demonstrating the position and shape of defects, and achieved a detection ratio of 97.37% on 342 experimental sample images, bearing a sound selfadaptability and robustness.

关 键 词：织物疵点检测 单演小波分析 RIESZ变换 拉普拉斯小波 多重调和样条 

分 类 号：TS101.9[轻工技术与工程—纺织工程]