MLMCD: A Machine Learning-Based Multi-Level Multi-Channel Denoising Framework for Robust Multi-Noise Image Restoration

 Image denoising is a core problem in digital image processing and plays a vital role 
in applications including medical imaging, remote sensing, video surveillance, and 
autonomous systems. While numerous denoising techniques have been developed, most are 
designed for specific noise models such as Gaussian, impulse, or speckle noise, resulting in 
significant performance degradation under mixed and high-density noise conditions 
commonly encountered in real-world imaging scenarios. 
To address these limitations, this paper proposes MLMCD (Multi-Level Multi-Channel 
Denoising), a hierarchical filtering framework that integrates multi-scale spatial 
decomposition with cross-channel information fusion. The proposed multi-level architecture 
progressively refines noise estimates across resolution stages, enabling effective coarse-level 
noise suppression while preserving fine structural details at higher resolutions. In addition, a 
dedicated multi-channel fusion module exploits inter-channel correlations in colour images to 
achieve coherent noise attenuation across RGB channels without compromising chromatic 
fidelity. An adaptive noise estimation mechanism dynamically regulates filtering strength 
based on locally computed noise statistics, removing the need for explicit noise-type 
specification and enhancing robustness to previously unseen mixed-noise conditions.