基于傅里叶特征谱和相关系数的织物疵点检测

Fabric defect detection using characteristic spectrum of Fourier transform and correlation coefficient

为提高织物疵点自动检测的准确度,提出一种基于傅里叶特征谱和相关系数的织物疵点检测算法。以平纹、斜纹织物为研究对象,对织物图像进行傅里叶变换,得到织物图像的频谱图;定位频谱中的特征峰点,提取表征图像灰度、纹理的五个特征值;以正常织物为模板,计算待检图像特征值与模板图像特征值之间的相关系数,确定用于识别织物疵点的阈值,来实现织物疵点检测。实验结果表明:当阈值设定为0.80时,该算法能够实现稀密路、断经、吊经、纬缩、破洞等常见疵点的准确检测。

A detection algorithm based on characteristic spectrum of Fourier transform and correlation coefficient is put forward to improve the accuracy of automatic fabric defect detection. With plain weave and twill fabrics as the researched objects, the frequency spectrum diagram of fabric image can be obtained by Fourier transform. The characteristic peaks in the spectrum are located and five characteristic parameters of image gray and texture are extracted. Correlation coefficients between characteristic parameters of being detected images and the template image are calculated using the faultless fabric as a template, and then the threshold value of defect detection is determined to realize the defect detection. Experimental results show that the algorithm proposed in this paper can achieve accurate detection of common defect, such as weft crackiness, cracked ends, stretched warp, looped weft, holes and so on, when the threshold is set as 0.80.