High-fidelity dual-wavelength phase retrieval for lensless wavelength-multiplexed microscopy

Lensless wavelength-multiplexed microscopy provides an effective imaging solution for single-frame wavefield recovery, in which three lasers with different wavelengths simultaneously illuminate a sample to capture a color hologram. In data processing, the recorded color hologram is divided into three intensity images and then input into multi-wavelength phase retrieval (MWPR) for image recovery. However, the performance of MWPR on data-reduced measurements, such as dual-wavelength imaging, is undermined by the stagnant convergence. In this work, we propose a new, to our knowledge, dual-wavelength phase retrieval to stay away from the local minimum and realize high-fidelity image recovery. In our design, the dual-wavelength phase retrieval is constructed into four steps, including alternative projection, total variation regularization, structural patch decomposition, and image denoising, to form an iterative optimized process to reconstruct the sample image. Experimental results demonstrate the effectiveness and superiority of our method in terms of laser-based and LED-based imaging systems.