DEALing with Image Reconstruction: Deep Attentive Least Squares

TL;DR

This paper introduces a novel image reconstruction method that combines traditional regularization principles with deep learning techniques, using learned filters and attention mechanisms for improved interpretability and robustness.

Contribution

It presents a data-driven iterative reconstruction approach inspired by Tikhonov regularization, integrating learned features and attention for enhanced performance and interpretability.

Findings

Achieves performance comparable to state-of-the-art methods

Offers improved interpretability and robustness

Demonstrates convergence and principled reconstruction

Abstract

State-of-the-art image reconstruction often relies on complex, highly parameterized deep architectures. We propose an alternative: a data-driven reconstruction method inspired by the classic Tikhonov regularization. Our approach iteratively refines intermediate reconstructions by solving a sequence of quadratic problems. These updates have two key components: (i) learned filters to extract salient image features, and (ii) an attention mechanism that locally adjusts the penalty of filter responses. Our method achieves performance on par with leading plug-and-play and learned regularizer approaches while offering interpretability, robustness, and convergent behavior. In effect, we bridge traditional regularization and deep learning with a principled reconstruction approach.