The Research Of Fabric Defects Of Coarse Positioning And Detection Algorithm

In the textile production, the fabric defect detection directly affects the quality of thetextile. With the rapid development of the computer and image processing technologies, thefabric defect detection has been paid a great attention.Based on NMI extracting features in the gesture recognition, this paper proposes a coarselocalization algorithm of detecting fabric defects. Firstly, we de-noise for fabric defect images,then do binarization. Secondly, according to the shapes of defects, we divide the binarizedimage into some sub-blocks, and calculate the corresponding NMI of every sub-block. Lastly,we extract the sub-blocks that NMIs are not equal to0, thereby, we achieve the coarselocalization for defects. Experimental results show that the proposed algorithm reaches betterresults for three kinds of fabric defects.Moreover, since Gabor filters in the time-frequency domains have good local resolutioncapability, and the spectral specificity of textures as well as their variation with spatialpositions, therefore, Gabor filters have been widely applied in the fabric defect detection.Based on Gabor filters, this paper proposes two algorithms of the fabric defect detection:(1) At first, by using the half-peak tangent method, we design12circular-shaped realGabor filters, which distributed at3scales and4orientations. Such filters can cover thefrequency of defects as possible. Then, based on these filters, we propose a new algorithmdetecting fabric defects, in which we use the opening operation in the image morphological toremove noise. At last, we apply the new algorithm to detect defects for8fabric defect imageswith the same texture background, and the experimental results show that the new algorithmcan detect defects accurately.(2) In order to reduce the amount of computations, combining with a new method of theimage enhancement, we improve the algorithm in (1) and propose a new algorithm detectingfabric defects based on complex Gabor filters. Combining the new method of the imageenhancement with Hardmard product of matrices, it can effectively enhance the contrastbetween the defect and background. Relative to the algorithm in (1): the filters is reducedfrom12to8; in the convolution operation, when calculating a pixel value of some point, thepixel values involving in the operation are reduced from49to11; the fusion method is moresimpler. Based on the above three aspects, although we add the image enhancementoperation, the present algorithm greatly reduces the amount of computation. Experimental results show that the new algorithm can better detect the defects, and the details of defects isbetter detected, but the algorithm in (1) better handle the overall outline of defects.