结合自适应中值滤波和GGCM的织物质量检测

Fabric defect detection method based on adaptive median filtering algorithm and GGCM

在纺织品的生产过程中,缺陷检测不可或缺。其中,特征提取是关键步骤,而在特征提取前进行合适的预处理滤波可以有效提高特征对有缺陷样本和无缺陷样本的分离能力。文章提出一种结合自适应中值滤波和灰度-梯度共生矩阵(gray level-gradient co-occurrence matrix,GGCM)的织物质量检测方法。该方法使用灰度-梯度二阶统计特征来描述缺陷的纹理特征,并在预处理过程中利用自适应中值滤波来降低背景纹理信息对缺陷部分的影响,可有效提高纺织品的检测率。实验对含7种缺陷的纺织品样本进行GGCM特征提取,并输入到最近邻分类器中进行分类检测。与传统灰度共生矩阵方法的检测率(65.75%)相比,文中所提方法的检测率有所提高,达到了87.89%。实验结果表明,该方法对有缺陷的织物具有较高的分类能力。

Defect detection is very essential in the process of textile production. Feature extraction is the most important step of defect detection. The filtering before feature extraction is an effective measure to improve the characteristic of separation between defective samples and defect-free samples. In this paper, a new way of fabric defect detection combined with adaptive median filtering algorithm and gray level-gradient co-occurrence matrix（GGCM） is presented. In the method, the second-order statistical characteristic of GGCM is used to describe the texture characteristic of the defect, and the influence of the background texture information on the defect part is reduced by the adaptive median filter in pre-processing. The detection rate of textiles can be improved by the mixed processes. The GGCM feature of the fabric sample containing seven kinds of defects is extracted, and the detection rate is obtained by using the nearest neighbor classifier. The detection rate is 65. 75% by traditional GLCM method, but the proposed method can reach the detection rate of 87.89%. The experimental results show that the proposed method has high ability of classifying defect textures.