A Novel Patterned Fabric Defect Detection Algorithm based on Dual Norm Low Rank Decomposition

Fabric defect detection plays an important role in the quality control of textile manufacturing. In order to accurately detect the patterned fabric defects, a novel pattern fabric defect detection method based on the dual norm low rank decomposition is proposed in this paper. Firstly, the feature matrix is generated using Gabor transform. It makes the background lie in a low-rank subspace. Then, the dual norm low rank decomposition model is adopted to divide the feature matrix into low-rank part (background) and non-low-rank part (defect). The adopted model uses the dual norm of nuclear norm as the regular term to substitute for the original "sparse" constraint in the low rank sparse decomposition model, thus can realize the effective separation of the defect part by minimizing the inner product between background part and defect part. Finally, the saliency map generated by the non-low-rank part is segmented via the improved adaptive threshold to locate the defect regions. The experimental results show that the proposed algorithm has a high detection accuracy and robustness to various patterned fabric images comparing with the state-of-the-art.