<p>Infrared and visible image fusion (IVIF) aims to generate a single informative image by combining the complementary advantages of infrared and visible modalities. However, existing approaches often suffer from limited feature extraction capacity and insufficient preservation of thermal targets and fine structural details. To address these challenges, we propose a novel framework termed FDCFusion (Frequency-Domain CAFormer-CNN Fusion<b>)</b>, which leverages a frequency-domain dual-branch encoder architecture. Specifically, a Shared-Frequency Decomposition Module (SFDM) decomposes the source images into low-frequency and high-frequency components. The low-frequency components are fed into the CAFormer branch to model global semantic structures, while the high-frequency components are processed by the CNN branch to preserve fine-grained local textures. To achieve effective feature interaction, a Feature Integration Module (FIM) is further designed to adaptively integrate multifrequency representations through a multiscale channel integration strategy. Extensive experiments on publicly available datasets demonstrate that FDCFusion consistently outperforms six representative state-of-the-art fusion methods. Quantitatively, on the TNO dataset, FDCFusion improves SD, SF, SCD, and Q^ab/f by 1.66%, 6.27%, 3.76%, and 5.17%, respectively; on RoadScene, SD, SF, SCD, and Q^ab/f are increased by 2.12%, 5.63%, 3.07%, and 7.16%; and on MSRS, SD, MI, and Q^ab/f are enhanced by 1.58%, 3.57%, and 2.11%.</p>