Research On The Application Of Computer Vision In Fabric Defect Automatically Detect

Fabric defect detection is an important inspection item in textile quality control, and also the most difficult process to be automatized in the procedure of textile production. So far, fabric defect detection relies chiefly on visual detection, which not only has lower detection efficiency, higher false detection rate and omission rate, but also is labor-intensive and bad for worker’s health. So, it is of great importance to study technique for automatically detecting fabric defect based on computer vision.In this thesis, the situation of theoretical study and system development for automatically detecting fabric defect at home and abroad has been analyzed in a comparative all-round way. Based on this, a comparatively deep research has been done on key problems involved in fabric defect detection, such as image denoising pre-processing, defect area identification, defect’s texture feature extraction, and so on. The following will introduce the detailed step. Main research and results are as follows:1. The origin, type and characteristic of noise raised during the process of fabric image acquisition are analyzed. Combined with fabric’s texture feature, the thesis extraction, an image pre-processing method based on median filter and Laplacian image sharpening is presented.2. By use of the combination of selfadaptive threshold segmentation, mathematical morphology, and Canny edge detection algorithm, the thesis presents a rapid method for detecting monochrome fabric defect areas. First the image after preprocessing is segmented with adaptive threshold, so that to identify defects’ positions. Then, the mathematical morphology algorithm has been used to process those isolated spots and incomplete defect areas existing in the image after threshold segmentation, so as to preserve the connectedness of defect areas and remove irrelevant details. Thirdly, Canny edge detection algorithm is used to identify the edges of various defect areas and to obtain the information of defect areas’ positions, shapes, and sizes.3. Wavelet analysis has been used to decompose defect area image and to obtain warp sub-images and weft sub-images. By use of extracting those sub-images’ eigenvalue of energy, range, variance, entropy and inverse and normalizing them, it provides conditions for identifying defect types.4. Based on the foregoing theoretical and method research, the program for automatically detecting defects has been developed by use of OpenCV under Visual++6.0. A lot of experiments have been done under laboratory conditions on the common defects of monochromatic fabric, such as buckle-off, loom fly, scoring, holes, unclean color, gauze, rusty spot, hybrid fiber, smear and weave holes. The results verify that the foregoing method is feasible and effective.