基于动态聚类的织物疵点识别算法

Fabric defect detection based on dynamic clustering arithmetic

首先对织物图像进行高斯模板平滑预处理,然后提取正常织物图像小幅窗口的行列(对应织物中经纬向)灰度均值,再利用K均值聚类算法提取织物图像中纱线条干的灰度聚类中心和纱线空隙部分的灰度聚类中心,计算出正常织物图像行列灰度均值与其灰度聚类中心的波动范围。比较待检图像的行列灰度均值与正常织物图像的行列灰度聚类中心,判断其波动是否超出正常织物图像的波动范围,以判别待检图像中是否含有疵点,并正确标出织物图像中疵点位置。

The fabric image was processed with Gauss template smoothing pretreatment, and average grey value of columns and rows of the normal fabric image in small window was extracted. Mid grey value of the yarn evenness and the space between yams were worked out through K-means clustering. Besides, average grey value and fluctuating range of mid grey value of the normal fabric image were also figured out. By comparing mid grey value of the image being checked and of normal fabric image, whether the fluctuating range of the former went beyond that of normal fabric image was judged, and whether there were flaws in the image being checked was distinguished and its location was lined out exactly.