The Design Of Real Gabor Filters And Their Applications In Fabric Defect Detection

The fabric defect detection is one of the most important parts of the quality control, and itis the most difficult to realize the automation in the textile production. The research on thefabric defect detection can just contribute to the realization of the automation. Gabor filtersare one of the most useful tools to detect fabric defects, because Gabor filters have optimalresolutions in both the spatial and frequency domains, and the spectral specificity of texturesas well as their variation with spatial positions.In this paper, three kinds of real Gabor filters are designed, then the correspondingalgorithms are proposed to detect fabric defects, respectively.(1) Based on the features of common defects and the detection effect of fabric images byGabor filters,4ellipse-shaped real Gabor filters are designed. These filters are distributed athorizontal and vertical directions, and the magnitude contours of two filters in different scalesare tangent at every direction. Then a new algorithm based on these filters is proposed todetect fabric defects. The experimental results show that the new algorithm can detect manykinds of fabric defects accurately. Finally, the new algorithm is compared with other one inthree aspects: algorithm steps, computational complexity and experimental effect. The resultsindicate that the new algorithm has lower complexity and better effect.(2) In order to expand the detection range, firstly,8,12and16circular-shaped real Gaborfilters are designed. These filters are distributed at2,3,4scales and4orientations,respectively, and all the neighboring magnitude contours of the filters are tangent. Secondly, anew algorithm is proposed to detect fabric defects based on these filters, especially, a newmethod is presented in the step of fusing images. The experimental results show that the newalgorithm can detect many kinds of fabric defects accurately.(3) In order to expand the detection range and reduce computational complexity,circular-shaped real Gabor filters are designed by half-peak tangent method, in whichhalf-peak magnitude contours of the neighboring filters are tangent. On this basis, we proposea new algorithm to detect fabric defect, which includes the determining of optimal real Gaborfilters and detecting of fabric defects. Firstly, two optimal real Gabor filters are determined bydirection masks and Signal to Noise Ratio. Then the detailed operations are given to detectfabric defects. The experimental results show that the new algorithm can accurately detect thedefects not only for gray fabric defect images but also for color ones, and the true detection rate of the corresponding defect detection images is high.