MultiResolution Low-Rank Regularization of Dynamic Imaging Problems

TL;DR

This paper introduces a multi-resolution low-rank regularization method for dynamic imaging, leveraging wavelet transforms to improve image quality and reduce artifacts in dynamic X-ray tomography.

Contribution

It presents a novel multi-resolution low-rank decomposition approach that enhances regularization in dynamic imaging by applying local low-rank decomposition in wavelet domains.

Findings

Reduces block artifacts compared to traditional methods

Performs effectively in dynamic X-ray tomography

Comparable or improved image quality

Abstract

MultiResolution Low-Rank decomposition is formulated for regularization of dynamic image sequences. The decomposition applies a local low-rank decomposition on a sequence of discrete wavelet transforms. Its effective formulation as a regularization functional is discussed and numerically tested for dynamic X-ray tomography in comparison to other low-rank methods. The results suggest it is similar to traditional locally low-rank decomposition but produces less severe block artifacts.