用于织物疵点检测的最优Gabor滤波器设计

Design of Optimal Gabor Filters for Textile Defect Detection

提出一种基于织物纹理特征的最优Gabor滤波器设计方法.分别建立了正常纹理匹配和疵点纹理匹配的Gabor滤波器优化设计模型,并采用小生境遗传算法对两种模型进行求解.通过比较和分析两种滤波器的检测结果发现,由正常纹理匹配模型得到的最优Gabor滤波器更适宜于织物疵点的识别与分割,并且其中心频率与纹理图像功率谱中能量最集中的谐波成分相一致,因而可以极大地缩短求解优化模型所花费的时间.*

A design approach based on the characteristics of textile texture is developed for the optimal Gabor illters. Two types of optimization design models of Gabor filters which can match normal texture and defect texture respectively are formulated, and the niche genetic algorithm is employed to solve the models. By comparing and analyzing the detection results, it is found that the normal texture matched Gabor filters are more suitable than the defect texture matched Gabor filters, and the center frequency of the normal texture matched Gabor filters is consistent with the highest energy harmonic element in the power spectrum of the texture image, which greatly reduces time spent on solving the optimization model.