Texture-based Surface Segmentation Using Second-order Statistics of Illumination Series

Abstract: Many automated visual inspection applications rely on a segmentation of surfaces into meaningful regions, for instance into defective and non-defective areas. This paper presents a segmentation approach based on illumination series, by which we denote a set of images taken under variable directional illumination. We show that co-occurrence matrices calculated from the series of images enable the extraction of suitable features for a texture-based segmentation of the surface. Depending on the selected displacement vector, the co-occurrence matrices computed within a neighborhood contain information about spatial variations of the surface or about the average reflection properties. The method is developed on synthetic images and is then demonstrated with cutting inserts to segment areas featuring abrasion.