k-means聚类算法在织物疵点检测中的应用

Fabric defect detection based on k-means clustering

为检测常见织物的各种疵点,提出一种基于k-means聚类的织物疵点检测方法。对采集的图像进行中值滤波,以减轻纹理对疵点检测的影响,并利用方差采样算法增强织物的疵点特征信息;利用k-means聚类算法对方差采样后的图像进行处理,使得疵点区域被划分一类,非疵点区域划分为一类。最后经过二值化,分割出疵点。实验证明,该方法能快速、准确的检测出织物的常见疵点。与其他方法相比,文章提出的算法采用聚类思想对织物疵点进行分割,不需要利用正常织物图像进行阈值计算;另外经过方差采样算法处理后疵点信息明显增强,使得疵点信息与纹理明显不同,从而使聚类更为准确,增加了检测的准确度。

To detect a variety of common fabric defects, a new method based on k-means clustering is proposed. Firstly, the median filtering method is applied to reduce the impact of the texture on fabric defect detection, and the sampling variance algorithm is used to enhance the fabric defect feature information. Then, the k-means clustering algorithm is used to process the image after sampling variance, so that the image could be divided into two types of regions： defect area and non-defect area. Finally, defects are segmented using binary. Experimental results showed that the proposed method could detect the common fabric defects quickly and accurately. Compared with other methods, the proposed method uses clustering ideological to segment the fabric defect, and does not require calculated threshold. In addition, the defects information are obviously enhanced using the variance sampling algorithm, so the defects information becomes very different from texture. Therefore the cluster of sample are more accurate, and the accuracy of detection are higher.