Two-frame randomly phase-shifted noisy interferograms phase retrieval based on intermediate frame generation network

Accurate phase retrieval from interferograms is crucial for interferometry. Traditional two-frame phase-shifted interferograms phase retrieval methods offer limited accuracy and fail to adequately account for physical processes and noise effects. In this paper, we propose what we believe to be a novel method for phase retrieval from two-frame randomly phase-shifted noisy interferograms. The intermediate frame generation network (IFGNet) takes two-frame interferograms as input and generates five output frames: the denoised versions of the input interferograms and three intermediate frames that represent the phase-shifting process. The network architecture incorporates a multi-channel attention mechanism, which effectively leverages the physical information within the interferograms, enabling precise, noise-free interferogram prediction. Phase retrieval is ultimately achieved using the Stoilov method applied to the five output interferograms. The application of the target-based training strategy (TBTS) further enhances the accuracy of the results. Through simulations and experiments, our method outperforms existing two-frame phase retrieval techniques in terms of accuracy and robustness, demonstrating its potential for practical applications.