A NEURAL NETWORK APPROACH TO THE PHASE UNWRAPPING PROBLEM IN FRINGE ANALYSIS

This paper presents a novel approach to the phase unwrapping problem by employing a back-propagation neural network to detect the presence of phase wraps in an image. The philosophy behind the approach is to keep the analysis simple by using a small network consisting only of six input. six hidden and six output neurons. Each input neuron is assigned to one pixel and this input "window" is convolved with an image to analyse only six pixels at a time. The unwrapped phase distribution is reconstructed fromthis series of analyses. It is shown that after training for approximately two hours. the network can successfully unwrap a one-dimensional phase distribution in 0.5 seconds and that this method could prove to be the basis for a robust two dimensional phase unwrapper.