This is a Tensorflow implementation of the SPL2020 paper A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network
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.
python main.py
- One sample data in fastMRI dataset named 'MRI_Knee_58.mat' is in folder './data/'
- The settings of plug-and-play algorithm and neural network are in file settings.py
- The MSSN is trained on BSD500 dataset, and the checkpoints are in folder './models/checkpoints/'
- 36- and 48-line radial sampling of k-space are used in our experiments.
@article{song2019new,
title={A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network},
author={Song, Guangxiao and Sun, Yu and Liu, Jiaming and Wang, Zhijie and Kamilov, Ulugbek S},
journal={IEEE Signal Processing Letters},
year={2020},
doi={10.1109/TCI.2019.2893568},
}