Learning Iteration-Dependent Denoisers for Model-Consistent Compressive Sensing

Pavan Kumar Reddy, K and Chaudhury, KN (2019) Learning Iteration-Dependent Denoisers for Model-Consistent Compressive Sensing. In: 26th IEEE International Conference on Image Processing, ICIP 2019, 22 - 25 September 2019, Taipei, pp. 2090-2094.

PDF ICIP_2019.pdf - Published Version Restricted to Registered users only Download (2MB) | Request a copy

Abstract

Modern regularization techniques and iterative solvers have largely been the key to the success of Compressive Sensing (CS). Recently, deep neural networks (DNNs) with end-to-end training have shown promise for CS. However, because of their open-ended nature, it is difficult to ensure that the DNN output is consistent with the measurements. In contrast, iterative algorithms such as FISTA explicitly make use of the measurement model and are hence able to incorporate consistency. To strike a middle path, researchers have shown that the performance of traditional iterative solvers can be improved by formally replacing the proximal map at each iteration with powerful DNN denoisers. While existing denoisers are typically designed to handle additive white noise, the noise that the denoiser encounters during each iteration is highly correlated and difficult to characterize. Motivated by this observation, we propose to use iteration-dependent denoisers within the FISTA framework, i.e., we train separate DNNs that can specifically handle the noise encountered in the first few iterations. We are able to achieve state-of-the-art CS results with fewer iterations as result, while maintaining measurement consistency. © 2019 IEEE.

Item Type:	Conference Paper
Publication:	Proceedings - International Conference on Image Processing, ICIP
Publisher:	IEEE Computer Society
Additional Information:	The copyright for this article belongs to IEEE Computer Society.
Keywords:	compressive sensing; consistency; deep neural networks; denoising; noise model
Department/Centre:	Division of Electrical Sciences > Electrical Engineering
Date Deposited:	06 Jan 2023 06:48
Last Modified:	06 Jan 2023 06:48
URI:	https://eprints.iisc.ac.in/id/eprint/78814

Actions (login required)

View Item