Hybrid Deep Learning for Image Denoising: Balancing Noise Removal and Detail Preservation with Game Theory

Image denoising aims to suppress noise while preserving fine details, a challenge for multimedia and computer vision applications. Conventional convolutional models achieve effective denoising but often oversmooth textures.  Recently, Multi-Layer Perceptron (MLP) based structures have shown strong potential for image restoration.  In this work, we propose a hybrid framework that combines an MLP-based denoising module with a UNet-based detail-preservation network.  The interaction between the two components is formulated within a game-theoretic framework, where noise reduction and structural fidelity are modeled as competing objectives seeking a Nash equilibrium.  Experiments on the Smartphone Image Denoising Dataset (SIDD) demonstrate improved PSNR and SSIM compared to single-model baselines. Visual results confirm sharper edges and richer textures with effective noise suppression.  This study highlights the promise of combining MLP-based architectures and game theory to achieve robust and detail-preserving image denoising.