Year: 2026
Type: preprint
Abstract: Infrared and visible image fusion aims to integrate thermal radiation information and scene texture details into a single comprehensive representation. Existing deep learning methods typically encode the two modalities through separate network branches and perform fusion only at the feature level, overlooking the potential benefits of cross-modal interaction during encoding. Inspired by state space models, this paper proposes CMS 2 -Fusion, which introduces a cross-modal state space coupling mechanism that allows the hidden states of both modalities to dynamically interact within the Mamba block through a learnable coupling coefficient, modeling thermal-texture dependencies at the state-transition level. On the MSRS dataset, CMS 2 -Fusion achieves a Q abf of 0.7704 and an SSIM of 0.6562, surpassing classical CNN methods and a decoupled Mamba baseline, while reducing parameters by 13% (8.4M vs. 9.6M) through shared SSM parameters. This work is the first to apply state-space cross-modal interaction to infrared-visible fusion, realizing a paradigm shift from “encode-then-fuse” to “encoding-as-coupling.”
Source: Research Square
Authors: Jiajia Liu, Hanfei Zhang
Institution: Huaiyin Normal University
Language: English
Subfield: Media Technology
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225917537
