Automatic Visual Defect Detection Using Texture Prior and Low-Rank Representation

IEEE Access

Hong Zhang¹  Wenwei Huang²  Xiangrong Zeng³  Qizi Huangpeng³ 

[1] Department of Computing Science, University of Alberta, Edmonton, Canada;Department of Military Basic Education, National University of Defense Technology, Changsha, China;Department of Systems Engineering, National University of Defense Technology, Changsha, China;

关键词: Defect detection;    low-rank representation;    texture prior;   

DOI  :  10.1109/ACCESS.2018.2852663

来源: DOAJ

【 摘 要 】

Automatic surface detection for quality control has largely employed image processing techniques, for example in steel and fabric defect inspection. There are rising demands in the quality control industry for defective image analysis to fulfill its vital role in visual inspection. In this paper, we introduce an unsupervised method using a low-rank representation based on texture prior for detection of defects on natural surfaces and formulate the detection process as a novel weighted low-rank reconstruction model. The first step of the proposed method estimates the texture prior to a given image by constructing a texture prior map where higher values indicate a higher probability of abnormality. The second step of the proposed method detects the defect via low-rank decomposition with the help of the texture prior. Experiments on synthetic and real images show that the proposed method is superior in terms of detection accuracy and competitive in computational efficiency with respect to the state-of-the-art methods in surface defect detection research. This contribution is of particular interest for manufacturers (e.g., steel and fabric) for which defect detection largely relies on manual inspection.

【 授权许可】

Unknown