Autonomous Band Registration for On-Board Applications

Automatic band-to-band registration of multispectral images is the most critical pre-processing requirement for imaging tasks performed on board satellites, such as data fusion, intra-band coding, change detection, spectral signature based classification, etc. The performance of these tasks depends on the robustness and accuracy of band registration. The a priori information of band misalignment is not accurate enough, due to spacecraft shaking or attitude changes. This paper proposes an autonomous band registration system, which uses a new gradient-based phase correlation method with added sub-pixel extension stage. A new automatic determination scheme for valid registration is also proposed. Experimental results on natural satellite images show that the proposed autonomous system has higher accuracy and better robustness.