This paper is concerned with inspecting surfaces using the textural properties of the surface. The approach taken here is that of modelling the surface texture by a "stochastic" model which is parametric, synthesis, compact and parsimonious. Two such models are discussed: the Markov Random Fields and the Fractal models. The first model is very useful for modelling textured surfaces such as textile, lumber, etc; whereas the second one is useful for modelling perceptual surface roughness. Surface inspection is cast as a statistical classification and hypothesis testing problem based on the maximum likelihood estimate (MLE) of the model parameters (or on the sufficient statistics). The image is divided into disjoint square windows and a MLE a* (or a sufficient statistic) is computed for each window . A Mahalanobis metric 11 a* - a 11,p weighted by the Fisher information matrix 'P is computed and compared to a predetermined threshold. This metric is shown to be the quadratic of the likelihood of the data for a large number of samples, and the test is the corresponding chi-square test.

Date Published: 23 October 1986
PDF: 7 pages
Proc. SPIE 0665, Optical Techniques for Industrial Inspection, (23 October 1986); doi: 10.1117/12.938773

Published in SPIE Proceedings Vol. 0665:
Optical Techniques for Industrial Inspection
Paolo G. Cielo, Editor(s)