Defect inspection for structural texture surface based on generalized low-rank approximations of matrices

Abstract

© 2017 IEEE. In this paper, we propose an automated surface inspection approach based on generalized low-rank approximations of matrices(GLRAM). The GLRAM uses a series of low-rank matrices to approximate the structural texture backgrounds of the original defective images. Then the difference images obtained by the original defective images and the low-rank approximations will retain the defect areas. Finally, we can easily segment the defect area from the original image using a simple threshold segmentation method. Experimental results show that our proposed method can extract defect areas well from the structurally textured image.