Novel phase retrieval based on deep learning for fringe projection profilometry by only using one single fringe

Fringe projection profilometry (FPP) has become increasingly important for 3-D shape measurement because of its attributes of high-resolution, high-accuracy, and high-speed, etc. In the FPP, a phase retrieval process is necessary to retrieve the desired phase before the 3-D shape reconstruction, which usually includes two steps of phase calculation and phase unwrapping. Traditional techniques always require more than one fringe for successful phase retrieval, which is difficult to be used for dynamic 3-D measurement. In this paper, a novel phase retrieval technique based on deep learning is proposed by only using one single fringe, and the desired phase can be successfully retrieved by using the deep learning with one single network. The proposed phase retrieval technique shows great potential for dynamic 3-D measurement. Theoretical analysis and experiments are provided to verify its performance.