Multi-component separation, inpainting and denoising with recovery guarantees

TL;DR

This paper introduces algorithms for separating multiple geometric image components and inpainting missing regions, supported by theoretical guarantees based on compressed sensing and sparse representations, extending previous methods to general frames.

Contribution

It proposes novel algorithms for multi-component separation and inpainting with theoretical recovery guarantees using compressed sensing and extends sparse representation theory to general frames.

Findings

Successful separation of point and curvilinear singularities

Effective inpainting of missing image regions

Theoretical guarantees for algorithm performance

Abstract

In image processing, problems of separation and reconstruction of missing pixels from incomplete digital images have been far more advanced in past decades. Many empirical results have produced very good results, however, providing a theoretical analysis for the success of algorithms is not an easy task, especially, for inpainting and separating multi-component signals. In this paper, we propose two main algorithms based on $l_1$ constrained and unconstrained minimization for separating $N$ distinct geometric components and simultaneously filling-in the missing part of the observed image. We then present a theoretical guarantee for these algorithms using compressed sensing technique, which is based on a principle that each component can be sparsely represented by a suitably chosen dictionary. Those sparsifying systems are extended to the case of general frames instead of Parseval frames which have been typically used in the past. We finally prove that the method does indeed succeed in separating point singularities from curvilinear singularities and texture as well as inpainting the missing band contained in curvilinear singularities and texture.