2D Phase Unwrapping Based on U-Net and SVD for Fringe Projection Profilometry

Phase unwrapping, a vital step with challenges in the Fringe Projection Profilometry (FPP), is directed to retrieve the origin phases from the wrapped phases to finish 3D reconstruction of objects. But Noise and spectrum aliasing from image capturing will lead to error diffusion in traditional phase unwrapping algorithms. This paper proposes a new, fast and robust 2D phase unwrapping method (called SVD-UNet) based on Singular Value Decomposition (SVD) and semantic segmentation model U-Net. We equivalently convert the question of unwrapping phase into a classification issue of the order number of phase unwrapping. According to the features of the wrapped phase, it is easier for our method to obtain global information and improve the effect of semantic segmentation when taking the results of linear decomposition through SVD as the input of U-Net. The experimental results show that the new method has more stable performance, faster convergence speed and is more robust than the state-of-the-art.