Efficient Pansharpening Using Schrödinger Bridge Matching

Pansharpening can be formulated as an inverse problem and expressed as a unified degradation-recovery stochastic differential equation (SDE) or ordinary differential equation (ODE). By leveraging the Schrödinger Bridge (SB) formulation, the proposed method improves the degradation-recovery SDE/ODE to a linear forward-backward SDE/ODE, which exhibits excellent properties of optimal transport and enables efficient training and sampling.