Preliminary version of the paper: http://arxiv.org/abs/1803.02112

Abstract

We introduce a paradigm for nonlocal sparsity reinforced deep convolutional neural network denoising. It is a combination of a local multiscale denoising by a convolutional neural network (CNN) based denoiser and a nonlocal denoising based on a nonlocal filter (NLF) exploiting the mutual similarities between groups of patches. CNN models are leveraged with noise levels that progressively decrease at every iteration of our framework, while their output is regularized by a nonlocal prior implicit within the NLF. Unlike complicated neural networks that embed the nonlocality prior within the layers of the network, our framework is modular, it uses standard pre-trained CNNs together with standard nonlocal filters. An instance of the proposed framework, called NN3D, is evaluated over large grayscale image datasets showing state-of-the-art performance.

Software

The following software is released under TUT limited license. It can only be used for non-profit education and scientific research. Any unauthorized use of the software for industrial or profit-oriented activities is expressively prohibited.

Download

You can find the usage instructions enclosed.

The current software version is compatible with Matlab 2011a 64 bit, or more recent, for both Windows and Linux.

Acknowledgements

This work is supported by the Academy of Finland (projects no.287150, 2015-2019, and no.310779, 2017-2021) and European Union’s H2020 Framework Programme (H2020-MSCA-ITN-2014) under grant agreement no.~642685 MacSeNet.

The binary components GroupProcessor and BlockMatch present in the distribution were provided by the authors of BM3D.