Efficiently state-space model with cross-guided fusion for infrared and visible images

Current deep learning-based image fusion methods often face a trade-off between global receptive fields and computational efficiency, limiting their practical applications. To address this challenge, we propose a novel infrared and visible image fusion framework, CSMFusion. We design a feature extraction module based on the Residue Multi-Head SSM Block to model long-range dependencies and effectively extract local features. Residual connections are employed to preserve shallow information and reduce the loss of texture details. Additionally, we introduce a Cross-Gating SSM Module to facilitate cross-modal feature fusion, where the contribution of each modality is computed to fully exploit domain-specific information and achieve cross-domain complementary information integration. The fusion block aggregates both shallow and deep features to generate the final fused images. Experimental results on three distinct datasets demonstrate that the proposed CSMFusion excels in image fusion tasks, offering significant advantages in both performance and efficiency compared to existing methods, while also showcasing robust generalization ability in advanced visual tasks such as object detection.