Multi-Source Image Fusion

Multi-source image fusion studies how to combine images of the same scene captured by different sensors, at different exposures or focus settings, or at different resolutions. Methods range from multiscale and frequency-based processing to CNNs, transformers, diffusion models, and self-supervised or unpaired learning; registration and degradation-aware processing help when inputs are misaligned or noisy. Applications include infrared-visible perception, multi-focus and HDR imaging, and hyperspectral or pansharpened remote sensing, with evaluation spanning image-quality metrics and downstream tasks such as detection, segmentation, and localization. A central challenge is balancing detail and spectral fidelity with semantic usefulness and robustness, especially when modalities disagree, data are scarce, or benchmark scores do not match human preferences.