DFOSfusion: Infrared and visible image fusion based on multi-scale dynamic feature optimization

With the rapid advancement of deep learning technology, significant progress has been made in image fusion, which is crucial for applications like surveillance, remote sensing, and medical imaging. However, most existing methods fail to fully extract and optimize multi-scale features, often lacking fine detail preservation, global consistency, and effective redundancy minimization, thus limiting overall visual quality. In this paper, we introduce DFOSfusion, an innovative image fusion framework specifically designed to address these challenges. DFOSfusion combines Multi-Scale Dense Blocks (MSD) with a Dynamic Feature Optimization System (DFOS), where the MSD module captures multi-scale local details, and the DFOS dynamically enhances global features via cross-layer spatial and channel optimization, effectively minimizing redundancy and promoting feature consistency. An adaptive texture-aware loss function further ensures that texture and structural details are consistently preserved during the fusion process. Extensive experiments reveal that DFOSfusion not only delivers superior visual quality but also consistently outperforms state-of-the-art methods, excelling in feature optimization and image coherence across multiple challenging benchmark datasets.