Resolution-Agnostic Lensless Imaging via Fourier Neural Operators

TL;DR

This paper introduces a Fourier Neural Operator framework for lensless imaging that improves reconstruction quality and enables resolution-agnostic inference, applicable to various global-PSF modalities.

Contribution

The authors develop a spectral-domain FNO approach that outperforms U-Net baselines and achieves resolution-agnostic reconstruction without retraining.

Findings

FNO improves PSNR by 2.14 dB over U-Net baseline.

Reconstructs higher-resolution images with less than 1 dB PSNR loss.

Framework applicable to other global-PSF imaging modalities.

Abstract

Lensless cameras based on thin diffusers offer a compact alternative to conventional refractive imaging but rely on computational reconstruction, since the diffuser's point spread function (PSF) globally multiplexes every scene point across the sensor. Here, we report a Fourier Neural Operator (FNO) framework for this reconstruction task. Because a linear shift-invariant forward model reduces to a pointwise multiplication in Fourier space, the spectral-domain kernel of an FNO layer is structurally aligned with the DiffuserCam inverse problem. Using a compact DiffuserCam prototype and a 25,000-image natural-scene dataset, our FNO improves upon a U-Net baseline of comparable parameter count by $2.14$~dB in PSNR and $0.11$ in SSIM. The same FNO, trained exclusively at $128 \times 128$, reconstructs $256 \times 256$ and $512 \times 512$ measurements with less than $1$~dB loss in PSNR and no retraining, demonstrating resolution-agnostic inference. The framework is directly applicable to other lensless modalities with global PSFs, such as multimode-fiber endoscopy.