Self‐supervised Image Denoising Algorithm Based on Improved Spatial Adaptive Network

Sheng Wang ； Hao Wu ； Zhongyang Jin

《World Scientific Research Journal》 9卷5期 (2023/05) Pp. 177-188

https://doi.org/10.6911/WSRJ.202305_9(5).0021

摘要

An improved self supervised image denoising algorithm is proposed to solve the problems of blurring details and excessive artifacts in the image denoised by common denoising algorithms. This method combines the Neighbor2Neighbor self‐monitoring training strategy with the spatial adaptive network SADNet, and uses the denoising network in the improved SADNet replacement training strategy to extract and process features. This method uses 7 × 7 size convolution layer extracts sufficient shallow feature information to improve the connectivity between feature mapping and codec structure, so as to enhance the network's ability to deal with different noise levels; Combined with single aggregation module and residual structure, intermediate feature extraction structure is designed to enhance feature aggregation to retain more detailed information; The RSAB structure is used to sample and weight the texture, detail and other features in the image, so as to restore the detail information and reduce the image artifacts. The experimental results show that the PSNR and SSIM values of the proposed algorithm are the highest compared with the common algorithms such as DnCNN, FDnCNN and FFDNet when tested on common BSD300, Kodak24, Set14, McMaster and Urban100 datasets; At the same time, the texture details of the denoised image are the clearest, the edge contour is obvious, and the visual effect is the best.

關鍵字

Image denoising ； Self‐supervised learning ； Spatial adaptive denoising network ； Detail recovery ； Residual structure

參考文獻

Liu B Z, Liu J B. Overview of Image Denoising Based on Deep Learning[J]. Journal of Physics: Conference Series,2019,1176(2).

Fan L, Zhang F, Fan H. Brief Review of Image Denoising Techniques[J]. Visual Computing for Industry, Biomedicine, and Art, 2019, 2(7):1-12.

Bhawna Goyal, Ayush Dogra, Sunil Agrawal, et al. Image denoising review: From classical to state-of-the-art approaches[J]. Information Fusion,2020,55(C).

Zhang H, Zhu Y, Zheng H. NAMF: A nonlocal adaptive mean filter for removal of salt-and-pepper noise[J]. Mathematical Problems in Engineering, 2021, 2021.

B V D S Sekhar, P V G D Prasad Reddy, S Venkataramana, et al. Image Denoising Using Novel Social Grouping Optimization Algorithm with Transform Domain Technique[J]. International Journal of Natural Computing Research (IJNCR),2019,8(4).