Recent work has shown the effectiveness of the plug-and-play priors (PnP) framework for regularized image reconstruction. However, the performance of PnP depends on the quality of image denoiser used as a prior. In this letter, we design a novel PnP denoising prior, called multiple self-similarity net (MSSN), based on the recurrent neural network (RNN) with self-similarity matching using multi-head attention mechanism. Unlike traditional neural net denoisers, MSSN exploits different types of relationships among non-local and repeating features to remove the noise in the input image. We numerically evaluate the performance of MSSN as a module within PnP for solving magnetic resonance (MR) image reconstruction. Experimental results show the stable convergence and excellent performance of MSSN for reconstructing images from highly compressive Fourier measurements.
Description
A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network - IEEE Journals & Magazine
%0 Journal Article
%1 9018286
%A Song, G.
%A Sun, Y.
%A Liu, J.
%A Wang, Z.
%A Kamilov, U. S.
%D 2020
%J IEEE Signal Processing Letters
%K bayesian priors readings
%P 1-1
%R 10.1109/LSP.2020.2977214
%T A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network
%U https://ieeexplore.ieee.org/document/9018286
%X Recent work has shown the effectiveness of the plug-and-play priors (PnP) framework for regularized image reconstruction. However, the performance of PnP depends on the quality of image denoiser used as a prior. In this letter, we design a novel PnP denoising prior, called multiple self-similarity net (MSSN), based on the recurrent neural network (RNN) with self-similarity matching using multi-head attention mechanism. Unlike traditional neural net denoisers, MSSN exploits different types of relationships among non-local and repeating features to remove the noise in the input image. We numerically evaluate the performance of MSSN as a module within PnP for solving magnetic resonance (MR) image reconstruction. Experimental results show the stable convergence and excellent performance of MSSN for reconstructing images from highly compressive Fourier measurements.
@article{9018286,
abstract = {Recent work has shown the effectiveness of the plug-and-play priors (PnP) framework for regularized image reconstruction. However, the performance of PnP depends on the quality of image denoiser used as a prior. In this letter, we design a novel PnP denoising prior, called multiple self-similarity net (MSSN), based on the recurrent neural network (RNN) with self-similarity matching using multi-head attention mechanism. Unlike traditional neural net denoisers, MSSN exploits different types of relationships among non-local and repeating features to remove the noise in the input image. We numerically evaluate the performance of MSSN as a module within PnP for solving magnetic resonance (MR) image reconstruction. Experimental results show the stable convergence and excellent performance of MSSN for reconstructing images from highly compressive Fourier measurements.},
added-at = {2020-03-08T17:35:51.000+0100},
author = {{Song}, G. and {Sun}, Y. and {Liu}, J. and {Wang}, Z. and {Kamilov}, U. S.},
biburl = {https://www.bibsonomy.org/bibtex/2e405975629f8e8ebf011794c82241336/kirk86},
description = {A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network - IEEE Journals & Magazine},
doi = {10.1109/LSP.2020.2977214},
interhash = {8199d452dcfed7a7c0e7f9377bd5b080},
intrahash = {e405975629f8e8ebf011794c82241336},
issn = {1558-2361},
journal = {IEEE Signal Processing Letters},
keywords = {bayesian priors readings},
pages = {1-1},
timestamp = {2020-03-08T17:35:51.000+0100},
title = {A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network},
url = {https://ieeexplore.ieee.org/document/9018286},
year = 2020
}