Applying backpropagation neural networks to gauging problems within fringe analysis

Abstract

Automatic fringe analysis is being successfully applied to visual inspection and surface measurement. This is largely due to the decreasing cost of powerful image processing systems allied to modern optics technology, a combination which has given the impetus for the development of new fringe pattern analysis algorithms. Prominent amongst these techniques are phase stepping and Fourier fringe analysis. This paper investigates the possibility of applying the backpropagation neural network paradigm to three gauging (or classification) fringe analysis applications: (1) to classify five spherical surfaces of differing radii from two sets of their simulated fringe patterns created from a standard fringe simulation technique with two different simulation conditions; (2) to classify five real objects with surfaces of different radii of curvature from fringe patterns produced under two different illumination conditions; and (3) to identify eggs according to their given grades from their respective fringe.