Non-Gaussian Noise Removal via Gaussian Denoisers with the Gray Level Indicator

Shi K, Guo Z (2023)

Publication Type: Journal article

Publication year: 2023

Journal

Journal of Mathematical Imaging and Vision

Abstract

This paper focuses on the problem of image restoration under non-Gaussian noise. Various gray level indicators have been proposed as effective tools for removing different types of non-Gaussian noise in the diffusion equation framework. However, the application of the method is limited since diffusion equations can not recover details and textures in images. In this paper, we consider the time discretization of the method and rewrite each step as the weighted average of the results obtained from the previous step and a mean filter. By replacing the mean filter with another Gaussian denoiser, we establish a general algorithm that extends any off-the-shelf Gaussian denoiser for removing different types of non-Gaussian noise. The weight (i.e., the gray level indicator) varies with the type of noise and is constructed based on the amplitude of the noise to estimate the degree to which each pixel in the noisy image is contaminated. It is shown that the commonly used Gaussian denoisers from classical filters to patch-based approaches are all valid. Experimental results demonstrate that the proposed algorithm with BM3D outperforms recent methods for image restoration under random-valued noise, Cauchy noise, and multiplicative noise, respectively.

Involved external institutions

China Jiliang University / China Institute of Metrology

China (CN) Harbin Institute of Technology (HIT) / 哈工大

China (CN)

How to cite

APA:

Shi, K., & Guo, Z. (2023). Non-Gaussian Noise Removal via Gaussian Denoisers with the Gray Level Indicator. Journal of Mathematical Imaging and Vision. https://dx.doi.org/10.1007/s10851-023-01148-9

MLA:

Shi, Kehan, and Zhichang Guo. "Non-Gaussian Noise Removal via Gaussian Denoisers with the Gray Level Indicator." Journal of Mathematical Imaging and Vision (2023).

BibTeX: Download