Fabric Defect Detection Based On Low-rank Decomposition Model

The goal of fabric defect detection is to locate a variety of common fabric surface defects.This is crucial for improving the quality of fabric products.Currently,most of this work is done manually.However,the results of manual detection are often affected by subjective factors,and the accuracy is low.On the contrary,the automatic fabric inspection technology based on computer vision can guarantee high detection accuracy while meeting the requirements of consistency and objectivity.Therefore,developing fast and accurate fabric defect detection methods has become a hot issue in the field of computer vision.This paper introduces low-rank decomposition model into fabric defect detection,and then the feature classification problem is transformed into a matrix decomposition problem.The main content includes:A fabric defect detection method based on low-rank decomposition and gradient information is proposed.In this method,a noise regularization is added to the traditional low-rank decomposition model to characterize the image noise part and enlarge the gaps between defective objects and background in feature space.Considering the structural characteristics of patterned fabric,gradient information is also introduced to guide matrix decomposition,which can constrain the regularization term adaptively according to the mutation degree of the current pixel point,so as to guide the matrix decomposition and reduce the noise misjudgment.This ensures the accuracy of detection results.A fabric defect detection method based on low-rank decomposition and structured graph algorithm is proposed.In this method,a structured graph algorithm is designed,which divides a fabric defect image into defect-free blocks Characterizing local features and the defect blocks destroying cycle according to the graphic characteristics.During the merger,an adaptive threshold is set up according to the number of the cycles contained in the current block to encourage the merger within lattice and discourage the merger of defective blocks with the surrounding defect-free blocks.Especially,a defect prior is calculated to guide the matrix decomposition,so that our model can weaken the defect-free regions and stand out the defective region in the sparse term.A fabric defect detection method based on low-rank matrix refactor and generalized convolution is proposed.In this method,the SVD operator is replaced by the auto encoder to optimize the low-rank terms of the traditional model to obtain more accurate low-rank textures.This allows our model to achieve a low-rank term closer to the background of the input image.Meanwhile,a difference matrix between the reconstructed image and the input image is constructed to rough estimate the location of the defect,where a generalized convolution is used to remove invalid points.To solve the global error caused by the reconstruction algorithm,a noise term is designed.