Regularization with Sparse Vector Fields: From Image Compression to TV-type Reconstruction

TL;DR

This paper proposes a new variational image compression method using sparse vector fields, inspired by PDE-based edge detection and inpainting, with connections to TV and TGV regularization.

Contribution

It introduces a novel regularization functional based on sparse vector fields, bridging image compression and PDE-inspired denoising techniques.

Findings

Effective compression of high-resolution images demonstrated

The model exhibits properties similar to TV and TGV regularizers

Potential for application in image denoising discussed

Abstract

This paper introduces a novel variational approach for image compression motivated by recent PDE-based approaches combining edge detection and Laplacian inpainting. The essential feature is to encode the image via a sparse vector field, ideally concentrating on a set of measure zero. An equivalent reformulation of the compression approach leads to a variational model resembling the ROF-model for image denoising, hence we further study the properties of the effective regularization functional introduced by the novel approach and discuss similarities to TV and TGV functionals. Moreover we computationally investigate the behaviour of the model with sparse vector fields for compression in particular for high resolution images and give an outlook towards denoising.