Surface grading using soft colour-texture descriptors

Abstract

This paper presents a new approach to the question of surface grading based on soft colour-texture descriptors and well known classifiers. These descriptors come from global image statistics computed in perceptually uniform colour spaces (CIE Lab or CIE Luv). The method has been extracted and validated using a statistical procedure based on experimental design and logistic regression. The method is not a new theoretical contribution, but we have found and demonstrate that a simple set of global statistics softly describing colour and texture properties, together with well-known classifiers, are powerful enough to meet stringent factory requirements for real-time and performance. These requirements are on-line inspection capability and 95% surface grading accuracy. The approach is also compared with two other methods in the surface grading literature; colour histograms [1] and centile-LBP [8]. This paper is an extension and in-depth development of ideas reported in a previous work [11]. © Springer-Verlag Berlin Heidelberg 2005.