Title: Deep denoiser prior and smoothed projection landweber inspired block-wise compressed sensing
Authors: Chun-mei Zong; Yue-qin Zhang; Qing-Shan Zhao
Addresses: Department of Computer Science and Technology, Xinzhou Teachers University, Xinzhou, Shanxi Province, 034000, China ' College of Computer Science and Technology, Taiyuan University of Technology, Taiyuan, Shanxi Province, 030024, China ' Department of Computer Science and Technology, Xinzhou Teachers University, Xinzhou, Shanxi Province, 034000, China
Abstract: How to use effective image prior to reconstruct high-quality images is a key problem in compressed sensing reconstruction. By introducing instantiation priors, traditional optimisation model-based compressed sensing reconstruction methods enjoy good structural analysis ability. To further improve the reconstruction quality, the optimisation model-based method is combined with deep learning to introduce a deep denoiser prior into BCS-SPL algorithm via a plug and play technique. Notably, the denoising operator is obtained by training a multi-scale residual network with data-driven discriminant learning method. Multi-scale network can extract different scale feature information about the image, and the introduced deep prior is beneficial for reconstructing high-quality images. Experimental results exhibit that the proposed method can effectively improve the image reconstruction quality without the expense of too much computational complexity.
Keywords: compressed sensing; deep learning; plug and play; deep denoiser prior; smoothed projection landweber; block-wise compressed sensing; deep denoiser; residual network.
DOI: 10.1504/IJCSM.2022.125909
International Journal of Computing Science and Mathematics, 2022 Vol.15 No.4, pp.347 - 358
Received: 01 Mar 2021
Accepted: 22 Mar 2021
Published online: 04 Oct 2022 *