Fabric Defect Detection Method Based on Sparse and Dense Mixed Low-rank Decomposition

Abstract

On account of the issue that there is severe noise in the detection of defects by the traditional low-rank decomposition defect detection method, in this paper, we present an efficient fabric defect detection approach which utilizes the sparse and dense decomposition on the base of low-rank representation. Firstly, the fabric image is uniformly segmented into image blocks. Each image block is spanned into a column vector, which is assembled to constitute the fabric image feature matrix. Then, the sparse and dense mixed low-rank decomposition model is constructed with the introduction of the F norm. The presented model is optimized by alternating direction multiplier method (ADMM) and augmented Lagrange multiplier (ALM), and the low rank array, dense matrix and sparse array are obtained. Finally, a thresholding segmentation approach is employed to detect the defect area by partitioning the salience map. Experimental results demonstrate that the proposed method achieves an efficient detection property, and it is superior to the current approaches.