Pushing The Limits of the Wiener Filter in Image Denoising

As modern image denoiser networks have grown in size, their reported performance in popular real noise benchmarks such as DND and SIDD have now long outperformed classic non-deep learning denoisers such as Wiener and Wavelet-based methods. In this paper, we propose to revisit the Wiener filter and r...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2023-03
Main Authors: Bled, Clément, Pitié, François
Format: Article
Language:English
Subjects:
Image filters
Wavelet analysis
Wiener filtering
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:As modern image denoiser networks have grown in size, their reported performance in popular real noise benchmarks such as DND and SIDD have now long outperformed classic non-deep learning denoisers such as Wiener and Wavelet-based methods. In this paper, we propose to revisit the Wiener filter and re-assess its potential performance. We show that carefully considering the implementation of the Wiener filter can yield similar performance to popular networks such as DnCNN.
ISSN:2331-8422