Image reconstruction enhancement via masked regularization

TL;DR

This paper introduces a masked regularization technique that improves the accuracy of edge-sparsity based image reconstructions, enabling near-perfect results with less data by focusing regularization away from edges.

Contribution

The paper proposes a novel masked regularization method that enhances suboptimal reconstructions by leveraging edge detection to preserve edge information and improve accuracy.

Findings

Enhanced reconstruction accuracy with fewer data.

Effective preservation of edge details in images.

Improved results across multiple initial methods.

Abstract

Image reconstruction based on an edge-sparsity assumption has become popular in recent years. Many methods of this type are capable of reconstructing nearly perfect edge-sparse images using limited data. In this paper, we present a method to improve the accuracy of a suboptimal image resulting from an edge-sparsity image reconstruction method when compressed sensing or empirical data requirements are not met. The method begins with an edge detection from an initial edge-sparsity based reconstruction. From this edge map, a mask matrix is created which allows us to regularize exclusively in regions away from edges. By accounting for the spatial distribution of the sparsity, our method preserves edge information and and furthermore enhances suboptimal reconstructions to be nearly perfect from fewer data than needed by the initial method. We present results for two phantom images using a variety of initial reconstruction methods.