Research On The Online Detection Algorithm For Fabric Defects

Fabric defect detection is one of the most important procedures effecting the manufacturing efficiency and quality in fabric industry. So far, most domestic factories accomplish this work by human vision. However, the inspection task done by human is time-consuming and labor-intensive. Moreover, the accuracy of the detection, which is liable to be effected by the inspector's experience, proficiency and some other subjective factors, is devoid of consistency and reliability. Thus it is desiderated to apply a fast and accurate online detection method for fabric defects to meet the needs of current market. As a prolonged functionality of human vision, computer vision has been collecting more and more favor in the application of industrial surface detection such as fabric defect detection with the advances of digital integration and digital image processing techniques.In this thesis, the state of arts of the research and application for fabric defect detection is firstly investigated, and then the whole design scheme with our research achievements is expounded. Afterwards, the real-time requirement is quantified by the restriction and targets of the factual application environment. Then, some typical defect detection methods are implemented offline to detect defects in real fabrics with their respective advantages and drawbacks compared, and their possibilities of being applied online along with the problems are analyzed in the successive part. According to the restriction of both real-time and accuracy performance, a method of Label Co-occurrence Matrix (LCM) is proposed as a candidate for fabric defect online detection. The comparison of the proposed method with other methods is made from the point of both real-time and accuracy performance aiming at different kinds of fabric detection. Further, by introducing the idea of fuzzy logic, a method of Fuzzy Label Co-occurrence Matrix is proposed as an improvement of LCM. Finally, problems within this method are summarized, and some amelioration ideas are given, and at last the perspective and direction of the future research are explored.As for the proposed method, image of the regular fabric is often comprehended as a regular texture image. However, slight fluctuation of gray-level of a pixel can not destroy the integrity of the whole texture. In the proposed LCM method, each pixel is classified into a certain tonal class according to their gray-levels within a predefined classification rule. The LCM with its features describes the statistical spatial interaction of those pixels. This thesis mainly formulates the definition of LCM, the conformation of the classification rules, the computation of features,and the selection of parameters. And with the ideology of fuzzy logic, a method of Fuzzy LCM is proposed successively. The detection results are illustrated and some concrete real-time performances are compared with other methods. It is testified practically that LCM with its features possesses more simplicity and better defect discriminability than many other methods, which makes the application of online detection for fabric defects probable.