Fabric Defect Detection Algorithm Based On Multi-feature Fusion And Low-rank Decomposition

Fabric defect detection plays an important role in the quality control of the textile production.However,a wide variety of fabric images,complex textures,different types of defects bring great challenges for fabric defect detection.The traditional existing fabric defect detection cannot deploy on the product line of low accuracy and low efficiency.Recently,the low-rank decomposition technology has been widely used in salient object detection and achieved good performance.However,a single feature cannot make the non-defect background lie in a low-rank subspace.In this thesis,we conducted thorough researches on the fabric defect detection algorithm based on multi feature fusion low rank decomposition.The main research results are as follows.(1)A fabric defect detection algorithm based on multi-channel feature matrices and joint low-rank representation is proposed.First,the second-order multi-channel features are extracted by modeling the retinal P-type ganglion cell coding to solve the intractable representation of complex texture fabric images.Then,joint low-rank representation model is constructed to divide the multiple feature matrices into low-rank matrices corresponding to the background and sparse matrices corresponding to the defects.Finally,the improved adaptive thresholding segmentation algorithm is utilized to segment the saliency map generated from the sparse matrices to locate the defect region.(2)A fabric defect detection algorithm based on multi-channel features and tensor low-rank decomposition is proposed.In order to characterize the direction information of fabric image,the second-order multi-channel feature matrices are extracted to form the feature tensor.Thereafter,an efficient tensor low-rank decomposition model is proposed to decompose the feature tensor into a low-rank tensor and a sparse tensor by the alternating direction method of multipliers according to the tensor recovery techniques.Finally,the saliency map generated by the sparse tensor part is segmented via the improved adaptive thresholding segmentation algorithm to locate the defective regions.(3)A fabric defect detection algorithm based on cascade low-rank decomposition is proposed.Firstly,textons features and Gabor features of fabric images are extracted;then a cascade low-rank decomposition model is constructed to combine the prior knowledge of Gabor directional features with the global texture features of textons to improve the detection results.Finally,an improved adaptive threshold segmentation algorithm is used to segment the saliency map to locate the defect regions.