Stability of Scattering Decoder For Nonlinear Diffractive Imaging
This work addresses the stability of an existing method for image reconstruction under multiple light scattering, which is an incremental contribution to the domain of computational imaging.
The paper tested the robustness of the Scattering Decoder (ScaDec) deep learning method for nonlinear diffractive imaging, finding that its performance remains stable across variations in permittivity contrasts, number of transmissions, and input signal-to-noise ratios in simulated datasets.
The problem of image reconstruction under multiple light scattering is usually formulated as a regularized non-convex optimization. A deep learning architecture, Scattering Decoder (ScaDec), was recently proposed to solve this problem in a purely data-driven fashion. The proposed method was shown to substantially outperform optimization-based baselines and achieve state-of-the-art results. In this paper, we thoroughly test the robustness of ScaDec to different permittivity contrasts, number of transmissions, and input signal-to-noise ratios. The results on high-fidelity simulated datasets show that the performance of ScaDec is stable in different settings.