STP-Fusion: A Rank-Preserving Cross-Resolution Alignment Framework for Infrared and Visible Image Fusion

Cross-resolution infrared and visible image fusion requires front-end alignment because the two modalities are often captured at different native resolutions. Conventional interpolation-based operators align the modalities through numerical expansion, but they can compress feature rank and weaken structural consistency before fusion. This paper proposes an alignment framework based on the Semi-Tensor Product (STP), which maps low-resolution infrared features into the visible topology under a constructed algebraic expansion. An Affine STP module is introduced to adapt the aligned representation in regions with weak infrared response. Experiments on the LLVIP dataset demonstrate that the proposed framework achieves more consistent structural fidelity and gradient preservation than standard front-end alignment baselines. The fusion results validate the effectiveness of the alignment mechanism in cross-resolution scenarios.