Computer Vision-based Fabric Defect Automatic Detection Technology

Fabric defect detection is an important step in the production process of textiles. At present, fabric defect detection mainly relies on visual inspection, its disadvantages are high false detection rate, high omission rate, low detection efficiency, heavy labor intensity, and harm to workers. Therefore, it's particularly significant to develop the technology to detect fabric defect automatically.In this paper, based on the analysis and the synthesis of existing fabric defect detection theories and methods, the method of fabric image pre-process and automatic identifying method based on wavelet analysis were researched deeply. The experimental results verified the feasibility and validity of the method.First of all, the noise source and noise characteristics of the fabric image and the denoising methods were analyzed. Directed towards the problem that the detail information was lost in the denoising process, the median filtering and edge sharpening had been used to preprocess the image, so as to improve the image visual effects, make the edge and outline of the object in the image stand out, and make it easy for computer to process and analyze the image.Secondly, in order to reduce the environment impact on defect identification and increase the speed of defect detection, the paper presents a window segmentation method based on gray. By contrast a window's gray mean to the whole image's gray mean, the window whose gray mean exceeds the setting threshold would be made a proliferation of jiugongge, and the new window formed in doing so would act as the area to be detected. Then, the wavelet analysis was used to extract feature values, which effectively reduced the amount of calculation to extract feature values and the number to calculate the window.Thirdly, automatic defect detection based on wavelet analysis was presented. Four feature value, energy, entropy, variance and range were extracted from the latitude and longitude sub-images processed by wavelet decomposition. After normalization, the responsiveness of the different feature values to defect was checked under the common size, and thus the existence of defects and their exact locations were determined.Finally, the software to detect fabric defect automatically was developed by use of LabVIEW. On a self-made experimental device, eight kinds of common defects, buttonhole, hybrid fiber, loom waste, hole, oil sands, anti-wire, dirt and residue were detected and analyzed. The experimental results showed that the method could accurately detect the existence of defects and determine their location faster. The researches provided theoretical foundation for the practical application of the fabric defect detection technology.