The fusion of infrared and visible images highlights the target while preserving detailed information, which helps to comprehensively capture the scene information. However, the existing methods continue to face challenges in managing the integration of global and local information, as well as enhancing the extraction of detailed image features, thus ultimately leading to constrained fusion outcomes. To enhance the fusion effect, this paper proposes a dual-branch residual attention-based infrared and visible image fusion network (TBRAFusion). The network utilizes two key modules, TransNext and the dual-branch residual attention (DBRA) module, which are used to process the input images in parallel to extract contrast and detail information. Additionally, an auxiliary function is incorporated into the loss function. Compared with mainstream fusion models, TBRAFusion achieves better fusion results and metrics through these improvements. The experimental results on the TNO dataset show that TBRAFusion improves the metrics in entropy (EN), spatial frequency (SF), sum ofcorrelation differences (SCD), and visual information fidelity (VIF) by 0.42 $\mathrm{\%}$, 4 $\mathrm{\%}$, 3.9 $\mathrm{\%}$, and 1.2 $\mathrm{\%}$, respectively. Tests on the MSRDS dataset show improvements of 1.7 $\mathrm{\%}$, 5.4 $\mathrm{\%}$, 9.6 $\mathrm{\%}$, and 4.9 $\mathrm{\%}$ in EN, standard deviation (SD), SF, and SCD, respectively.

The purpose of image fusion is to combine information from different source images to produce a comprehensively representative image. Traditional autoencoder architectures often struggle to effectively extract both unique and shared features from these image types. A novel two-stage three-branch feature extraction method (TSTBFuse) was proposed in the study, specialized for the fusion of infrared and visible images. The proposed architecture employed a three-branch encoder that separately captured infrared-specific thermal radiation features, visible-specific texture details, and shared structural information. A two-stage end-to-end training strategy was introduced: the first stage focused on reconstructing the original input images to preserve modality-specific information, while the second stage leveraged the learned representations to generate high-quality fused images. we designed a comprehensive loss function combining mean squared error (MSE), structural similarity index (SSIM), and gradient loss, ensuring both pixel-level accuracy and structural integrity. Extensive experiments on public datasets (TNO, MSRS and RoadScene) demonstrated that TSTBFuse consistently outperformed seven state-of-the-art methods in both subjective and objective evaluations. Furthermore, the method exhibited strong generalization capabilities, successfully extending to challenging tasks such as magnetic resonance imaging-computed tomography (MRI-CT) medical image fusion and red-green-blue (RGB)-infrared image fusion without retraining. The code is publicly available at: https://github.com/QXinYue/TSTBFuse.

This paper presents a method called MCADFusion, a feature decomposition technique specifically designed for the fusion of infrared and visible images, incorporating target radiance and detailed texture. MCADFusion employs an innovative two-branch architecture that effectively extracts and decomposes both local and global features from different source images, thereby enhancing the processing of image feature information. The method begins with a multi-scale feature extraction module and a reconstructor module to obtain local and global feature information from rich source images. Subsequently, the local and global features of different source images are decomposed using the the channel attention module (CAM) and the spatial attention module (SAM). Feature fusion is then performed through a two-channel attention merging method. Finally, image reconstruction is achieved using the restormer module. During the training phase, MCADFusion employs a two-stage strategy to optimize the network parameters, resulting in high-quality fused images. Experimental results demonstrate that MCADFusion surpasses existing techniques in both subjective visual evaluation and objective assessment on publicly available TNO and MSRS datasets, underscoring its superiority.