A Robust Phaseless Imaging Approach Using Reverse Kullback Leibler Divergence and Wirtinger Flow

Phaseless imaging is a prominent field of study in many imaging modalities. In practical applications, the phaseless measurements usually contain noise and outliers, limiting the reconstruction algorithms' performance. In this paper, we study the reverse Kullback-Leibler divergence (RKLD) as a robust objective function for mitigating the effects of corrupted measurements. We optimize RKLD using the state-of-the-art Wirtinger Flow (WF) algorithm for image reconstruction. Then, we numerically demonstrate the performance of RKLD using simulated coded diffraction measurements and a real optical dataset. Additionally, we compare RKLD's performance with other well-known loss functions used by the WF framework, namely, £2-intensity, Poisson, and £2-magnitude. Our simulation results demonstrate that RKLD-based optimization provides higher quality reconstructed images than the other loss functions in the presence of noise and outliers.