Automatic surface inspection for directional textures using nonnegative matrix factorization

Abstract

A global image restoration scheme using nonnegative matrix factorization (NMF) is proposed in this paper. This NMF-based image restoration scheme can be used for inspecting the defects in directional texture surfaces automatically. Decomposing the gray level of image pixels into an ensemble of row vectors, we first reduce the data set from original data space into a lower-dimensional NMF space. The repetitive and periodical primitives are well reconstructed by two lower-dimensional basis and weight matrices with nonnegative elements, named nonnegative matrix approximation (NMA). Then the local defects will be revealed by applying image subtraction between the original image and the NMA. As a consequence, the directional textures are eliminated, and only local defects are preserved if they initially are embedded in the surface. A supervised heuristic, elbow of residual curve rule, is devised which helps users to determine a proper basis space size of a specific image. Experiments on a variety of directional texture surfaces are given to demonstrate the effectiveness and robustness of the proposed method.