Research On Detection Methods Of Fabric Defects Based On Gray Level Co-occurrence Matrix And Visual Information

Influenced by the globalization of commodity economy,the textile industry has more and more demands on the quality of fabric.The quality control of the fabric defect has become a vital link in the production process of the textile industry.Traditional fabric defect recognition relies on human inspection.This method is in high labor cost and low detection accuracy.With the progress of computer technology,automatic recognition of fabric defect has gradually replaced human inspection,and become the trend of the industry.Although fabric defect automatic recognition research has made great achievements,most of the methods are under the influence of negative factors including defect categories,light change and noise.These problems make the fabric defect detection still a hot research subject.In this thesis,we primarily study the detection algorithm in the inspection of fabric defect,and investigate the recognitions for five kinds of common defects.The main contributions of this dissertation are as follows.(1)Due to the problems in defect detection that gray level co-occurrence matrix needs large amounts of calculation and texture information is lost by complicated sliding window design,a method of fabric defect detection based on gray level co-occurrence matrix using image block processing is proposed.This method uses operations of the first order grayscale statistics to optimize the process of the gray level co-occurrence matrix.Instead of sliding window,this method uses block sub-image to replace the original image.It can retain effectively a lot of essential information in the calculation of gray level co-occurrence matrix.At the same time,the method does not require prior knowledge learning and the performance of the sub-image is better than the sliding window,which contribute to little resources in calculation.It is easy to draw the characteristic curve and determine the location of the flaw when the texture feature is analyzed by the sub-image.Compared with the traditional gray level co-occurrence matrix,experiment results show that the proposed algorithm can quickly and efficiently detect fabric defects.(2)Due to the above detection method for defect detection only considering fabric texture information and ignoring the visual information,a method of fabric defect detection using improved visual saliency is proposed.This method sets entropy and energy features in gray level co-occurrence matrix as saliency characteristics of the bottom-up visual mechanism,and extracts the saliency characteristics of the top-down visual mechanism with super pixel segmentation and saliency filter.It fuses these features to visual saliency map by saliency calculation.The experiment results show that the rate of this method is better than fabric defect detection based on gray level co-occurrence matrix using image block processing.(3)In the VS2010 platform,this thesis adopts modularization thinking to design and implement the total modules of fabric defect detection prototype system,such as image processing,defect detection,defect location and so on.Among these functions,the defect detection module implements two detection methods proposed in this thesis.