Single Image Super-Resolution Based on Anisotropic Diffusion
Pascal Monasse
⚠ This is a preprint. It may change before it is accepted for publication.

Abstract

This paper details an unsupervised algorithm for image super-resolution based on a reaction-diffusion model. The diffusion is anisotropic, inspired by Perona-Malik diffusion. The reaction term favors compatibility with the low resolution input image. The implementation is simple and mainly relies on numerical schemes to approximate the required differential operators. The different possibilities present in the literature are analyzed and tested, leading to the choice of the best ones. Qualitative experiments illustrate the results on some images.

Download