Class-Aware Fully-Convolutional Gaussian and Poisson Denoising

TL;DR

This paper introduces a fully-convolutional neural network for image denoising that leverages class-awareness and multi-level features to improve performance across various noise types and levels.

Contribution

It presents a novel class-aware, fully-convolutional architecture that effectively exploits the denoising process's gradual nature, outperforming previous methods.

Findings

State-of-the-art results for Gaussian and Poisson noise

Class-awareness improves denoising quality and reduces artifacts

Multi-level feature utilization enhances texture recovery

Abstract

We propose a fully-convolutional neural-network architecture for image denoising which is simple yet powerful. Its structure allows to exploit the gradual nature of the denoising process, in which shallow layers handle local noise statistics, while deeper layers recover edges and enhance textures. Our method advances the state-of-the-art when trained for different noise levels and distributions (both Gaussian and Poisson). In addition, we show that making the denoiser class-aware by exploiting semantic class information boosts performance, enhances textures and reduces artifacts.