基于图元分割与Gabor滤波的织物瑕疵检测方法

为实现含复杂图案织物的自动化检测,提出基于图元分割与Gabor滤波的织物瑕疵检测方法,对具有复杂周期变化图案的织物进行检测。根据图像纹理的周期变化规律,确定图案单位模板大小,即包含一个周期图案的晶格。对图像进行自适应分割,并通过图元分割获得单元图像元素。通过Gabor滤波器生成特征的响应分布,获取理想的模板晶格,根据提出的投票程序,通过分析其特征向量的Manhattan根据距离图元晶格差异的分布来识别瑕疵图元晶格。实验结果表明:检测方法对星形和箱形图案的织物样本数据集上检测效果较好,显著提高了样本的查全率。

Fabric defect detection method based on primitive segmentation and Gabor filtering

In order to deal with the complex fabric defect detection with periodic variation pattern, a fabric defect detection method based on primitive segmentation and Gabor filtering was proposed. The template size of the pattern unit was determined according to the periodic variation of the image texture, i.e. a lattice containing a periodic pattern. The image was adaptively segmented, and the image elements of the smaller unit are obtained by the primitive segmentation. The response distribution given by the Gabor filter to the convolutional lattice produced an ideal template lattice. According to the proposed voting procedure, the lattice of the defect is identified by analyzing the distribution of the lattice differences represented by the Manhattan distance of the eigenvectors. Experiments show that the method has better detection effect on the fabric samples of star and box patterns, and significantly reducing the recall rate of the samples.