Edge-based texture measures for surface inspection

Abstract

Pietikainen and Rosenfeld (1982) introduced a class of texture measures based on first-order statistics derived from edges in an image. the objective of this paper is to evaluate the performance of these measures and some new edge-based texture measures using two different types of data sets: images taken from the Brodatz album and images from a practical wood surface inspection problem. The results obtained for edge-based measures are compared to those obtained by popular second-order texture measures and tonal features. The role of the classifier on the performance is also studied by comparing the results obtained for three parametric classifiers and for a nonparametric k-nearest neighbor classifier. The results indicate that edge-based approaches are very promising for surface inspection problems, because they are relatively simple to compute and have performed very well in experiments.<>