Infrared and Visible Image Fusion via Decoupled Color Space and Multi-Channel Consistency State-Space Modeling

Abstract

Infrared and visible image fusion (IVIF) plays a pivotal role in enhancing scene comprehension within complex and challenging environments. Despite rising research enthusiasm, most IVIF methods still rely on traditional RGB or YCbCr color spaces, where hue and brightness mix, causing color issues. Although new ideas and deep models emerge, they struggle to capture fine textures, model global semantics, and integrate multi–scale info, leading to blurred details in complex scenes. In this regard, this paper puts forward a novel network architecture that combines the HVI color space and a MCC-VSS (Multi-Channel Consistency Visual State Space) module to optimize the fusion performance. Specifically, we design a loss function grounded in the HVI color space to ensure color consistency, brightness adaptability, and structural fidelity. Additionally, we propose the Mamba-based MCC-VSS module for texture processing, global feature modeling, and multichannel consistency. Comprehensive experiments across three openly available benchmark datasets reveal that our approach markedly surpasses existing state-of-the-art techniques in both colour fidelity and texture preservation.