Optimized registration and adaptive fast inpainting algorithm for phase retrieval

In the field of quantitative phase microscopy, although binocular microscopy systems can simultaneously acquire two intensity images in a single exposure, the positional deviation of the image field of view reduces the accuracy of phase retrieval. To retrieve high-precision phase, we propose an optimized registration and adaptive fast inpainting algorithm to correct the inconsistency of the image field of view. First, in the feature matching stage, by optimizing the feature point matching strategy and introducing angle gating constraints, the accuracy and robustness of the correction are significantly improved. Second, in the image inpainting stage, the adaptive Criminisi inpainting algorithm is used to fill the texture and preserve the structure of the lost effective boundary information caused by registration. Simulation results show that compared with the comparison algorithms, the proposed method is significantly superior in registration accuracy, inpainting quality (with a peak signal-to-noise ratio and a structural similarity index measure improvements of 10.33 dB and 11.5%, respectively), and computational efficiency (with an improvement of 91.5%). In a real experiment with a microlens array, the relative error between the phase retrieval result of the effective edge information and the actual phase is only 1.8%, proving the effectiveness and correctness of the proposed method.