Iterative shrinkage thresholding-based anti-multi-noise compression perceptual image reconstruction network

• Published in: Signal, image and video processing Vol. 18; no. 5; pp. 4569 - 4578
• Main Authors: Xiang, Jianhong, Liang, Qiming, Xu, Hao, Wang, Linyu, Liu, Yang
• Format: Journal Article
• Language: English
• Published: London Springer London 01.07.2024; Springer Nature B.V
• Subjects: Computed tomography; Computer Imaging; Computer Science; Constraints; Feature extraction; Image compression; Image degradation; Image Processing and Computer Vision; Image quality; Image reconstruction; Interference; Medical imaging; Multimedia Information Systems; Noise levels; Original Paper; Pattern Recognition and Graphics; Signal to noise ratio; Signal,Image and Speech Processing; Vision; Deep learning; Attention mechanism; Compressed perception; Image anti-noise; Image reconstruction
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-024-03095-3

Telemedicine imaging services usually require wireless transmission of a large number of medical images MRI/CT, etc., in the network, which are subject to noise interference and block effect during transmission and compression, leading to degradation of image reconstruction quality, thus affecting diagnostic accuracy. In this paper, iterative shrinkage thresholding (ISTA)-based anti-noise compressive perceptual image reconstruction network is proposed to solve the problem. The network adopts a constrained sparse model, which incorporates both orthogonal and binary constraints of the sampling matrix into the network; in addition, the network adopts feature extraction subnetwork, parameter initialization subnetwork, and reconstruction anti-noise subnetwork for compressed perceptual image reconstruction, incorporates the channel attention mechanism, and proposes a hybrid network for anti-noise deblocking in the reconstruction anti-noise subnetwork. Experiments show that the maximum peak signal-to-noise ratio achievable by the network is in the range of 29.70–39.31 dB under different noise interference scenarios.