A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network

arXiv[cs:CS], 2019

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 the denoisers used as priors. 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.

Download arXiv here

Citation

G. Song, Y. Sun, J. Liu, and U. S. Kamilov, “A New Recurrent Plug-and-Play Prior Based on the Multiple Self-Similarity Network”, arXiv:1907.11793.