Locally-Supervised Global Image Restoration

TL;DR

This paper introduces a learning-based image reconstruction method that effectively handles fixed, deterministic sampling patterns by leveraging image invariances, achieving high-quality results with less ground truth data.

Contribution

It proposes a novel self-supervised framework for image reconstruction from fixed sampling patterns, overcoming limitations of traditional methods that require fully sampled data.

Findings

Achieves competitive or superior image reconstruction quality.

Requires significantly less ground truth data.

Validates effectiveness on photoacoustic microscopy upsampling.

Abstract

We address the problem of image reconstruction from incomplete measurements, encompassing both upsampling and inpainting, within a learning-based framework. Conventional supervised approaches require fully sampled ground truth data, while self-supervised methods allow incomplete ground truth but typically rely on random sampling that, in expectation, covers the entire image. In contrast, we consider fixed, deterministic sampling patterns with inherently incomplete coverage, even in expectation. To overcome this limitation, we exploit multiple invariances of the underlying image distribution, which theoretically allows us to achieve the same reconstruction performance as fully supervised approaches. We validate our method on optical-resolution image upsampling in photoacoustic microscopy (PAM), demonstrating competitive or superior results while requiring substantially less ground truth data.