Deep Neural Network Models for Image Denoising: DnCNN, GAN, and Unet

School of Computer Science and Technology, Zhejiang Normal University, Jinhua, Zhejiang, China

^* Corresponding author: liulu20040522@zjnu.edu.cn

Abstract

Image denoising is a fundamental task in the fields of image processing and computer vision, aiming to remove noise while preserving fine structural details. To address the limitations of traditional approaches that often struggle to balance noise suppression and detail preservation, this paper proposes two enhanced models based on the Deep Convolutional Neural Network for Denoising (DnCNN): DnCNN-GAN and DnCNN-UNet. The first incorporates a Generative Adversarial Network (GAN) to improve the perceptual quality and structural fidelity of denoised results, while the second leverages the multi-scale feature fusion capability of the U-Net architecture, achieving greater robustness and generalization under varying noise levels. For training, cropped patches from publicly available datasets were used, with Gaussian noise of random intensity added to simulate realistic scenarios. Experimental results demonstrate that both proposed models outperform the baseline DnCNN in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), validating their effectiveness and practical value for real-world image denoising applications.