A deep learning-based approach to solve the height-slope ambiguity in phase measuring deflectometry

Abstract

Phase measuring deflectometry (PMD) shows excellent performance in measuring the shape of specular surfaces. However, the existing height-slope ambiguity affects the measurement of PMD. In this paper, we propose a deep learning-based method to solve the ambiguity and reconstruct the specular surface. A neural network was built to reconstruct the surface. A set of simulation surfaces were generated using two-dimensional Legendre polynomials to train and test the network. A spherical-shaped surface was reconstructed in a simulation to evaluate the validity of the proposed method and compare our method to the state-of-the-art method. Experimental results demonstrate a 35.88% improvement in this method’s accuracy. Our method provides a great improvement in measurement accuracy.