Unregistered Spectral Image Fusion: Unmixing, Adversarial Learning, and Recoverability
Researchers have developed a new unsupervised framework for fusing unregistered hyperspectral and multispectral images. This method aims to improve both the spatial resolution of hyperspectral images and the spectral resolution of multispectral images. The approach combines spectral unmixing for multispectral super-resolution with adversarial learning in latent space for hyperspectral super-resolution. The paper also provides theoretical guarantees on the recoverability of the fused images, offering new insights into unregistered hyperspectral-multispectral fusion. AI