Defect detection on leather by oriented singularities

Abstract

This paper presents a system for leather inspection based upon visual textural properties of the material surface. Defects are isolated from the complex and not homogeneous background by analyzing their strongly oriented structure. The patterns to be analyzed are represented in an appropriate parameter space using an optimization approach: in this way a parameter vector is associated to each different textured region in the original image. Finally a filter process, based upon knowledge about the parameter vectors representing the leather without defects, detects and classifies any abnormality. Subject terms: leather inspection, texture analysis, texture classification, oriented structures, parameter space, vectorial bases.