Phase unwrapping from fringe projection profilometry using Fourier iterative attentional feature fusion Res-UNet

In fringe projection profilometry, phase unwrapping is an indispensable step as the quality of the continuous phase image is restored depends on the accuracy of the phase value calculation. Phase unwrapping methods based on Residual-UNet (Res-UNet) have been vastly used. However, the skip connection of the original Res-UNet simply sums the features which leads to insufficient generalization ability and limited feature fusion scale. And the original residual block has the problem of incomplete feature fusion, which makes it impossible to obtain complete phase information. In this paper, an advanced Fourier iterative attentional feature fusion (FIAFF) module is proposed and is integrated into the design of Res-UNet architecture, called FIAFF Res-UNet. And the iterative attentional feature fusion module is introduced to instead of the ordinary sum operation in residual block for the first time. And a new loss function is proposed by integrating the Smooth L1 loss with the MS-SSIM loss function. We compare the results yielded by our method with those by four other methods in the field of quantitative evaluation and visual effect. The experiment results show that the validity of the proposed method.