Regular Texture Defect Detection Based on Matched Gabor Filters
基于匹配Gabor滤波器的规则纹理缺陷检测方法

The surfaces of many industrial products are formed from the textural primitives by placement rules on the image field. It is an important task of automatic vision inspection to detect the defects in these regular textures. In the paper, two types of matched Gabor filters, that is normal texture matched Gabor filters and defect matched Gabor filters, are designed based on power spectrum of the regular normal texture and the abnormal texture. The normal texture matched Gabor filters, whose band pass area is positioned on the dominant frequency components of the normal texture, are designed to uplift the normal texture while suppress the defect texture. On the other hand, the defect matched Gabor filters, the band pass area of which guarantees the predominance of frequency components of the defect texture, have the effects of uplifting the defect texture while suppressing the normal texture. Taking advantage of the recent progress of study on principle of human vision, a novel adaptive filter design method is presented. Experiments on edge enhancement for defect detection using these matched Gabor filters have yielded satisfactory results, as far as both the precision and the speed are concerned.