• 24 March 2020
    • preprint
    • Published by HAL in HAL
Abstract
Non-local self-similarity and sparsity principles have proven to be powerful priors for natural imagemodeling. We propose a novel differentiable relaxation of joint sparsity that exploits both principlesand leads to a general framework for image restoration which is (1) trainable end to end, (2) fullyinterpretable, and (3) much more compact than competing deep learning architectures. We apply thisapproach to denoising, jpeg deblocking, and demosaicking, and show that, with as few as 100K parameters,its performance on several standard benchmarks is on par or better than state-of-the-art methods thatmay have an order of magnitude or more parameters.

This publication has 24 references indexed in Scilit: