Phase error compensation based on Tree-Net using deep learning

The nonlinear effect in phase shifting profilometry (PSP) is an essential source of phase error in 3D measurement. In this paper, we propose a universal phase error compensation method with a three-to-three deep learning framework (Tree-Net). Perfectly meeting the phase error compensation requirements, Tree-Net can construct six-step phase-shifting patterns from three-step. As a result, this compact network of fringe-to-fringe transformation has excellent performance when coping with different PSP systems after only one training. Experimental results demonstrate that the phase error can be reduced by about 90% in three-step PSP, which verified the effectiveness, universality, and robustness of the proposed method.