Research On Monochrome Cycle Pattern Fabric’s Defects Detection Algorithm

Due to the wide application of electronic technology and computer technology, international cotton textile technology develops towards the direction of high quality, high yield, automation and continuous. Instead of manual detection, fabric defect au-tomatic recognition has become an inevitable trend in textile industry development. At present, a lot of fabric defect detection algorithm has been put forward. Espe-cially for non-woven fabric, plain weave, twill, etc with no patterns, there have had detection system been put into the factory. While, for the fabric with patterns, it has some difficulty to study the detection algorithm because of its random texture. In this thesis, we study the monochrome cycle pattern fabric’s defects detection algorithm. It’s divided into three stages, image preprocessing, cycle divided and cycle defects detection.Considering the distortion problem of pattern fabric, distortion correction is car-ried on firstly. In order to remove the influence of uneven brightness, we partitioning the image to estimate the background gray scale and generate a background gray-scale extension image. And then we subtract the background gray-scale extension image from the original gray image to separate the foreground pattern and the background. According to the 2D Gabor filter’s frequency and direction close to the human visual system in the characteristics of frequency and direction, we construct a 2D Gabor filter to enhance the pattern, and do partitioning otsu binarization.Considering the characteristic of the pattern’s periodic arrangement, the cycle should be the value that the variance is minimum after the pattern is divided by the cy-cle. According to that, we get the pattern lateral cycle and obtain four cameras’lateral cycle in the initialization phase. Base on cycle, we complete the image cycle division and pattern templates.The cycle defects detection includes the foreground detection and background detection. Through the statistic of background gray scale, it automatically choos- es threshold to segment the abnormal grayscale defect area. To detect foreground pattern, we introduce 7 Hu invariant moments as pattern’s feature. Through fea-tures’similarity matching, it can pick out the defect pattern. The experimental results indicate that the proposed algorithm is feasible and effective.