New adaptive algorithms for real-time registration and fusion of multimodal imagery

Accurate image registration is a pre-requisite for most systems utilising two or more imaging sensors. This can often be accomplished off-line in the laboratory using appropriate test targets and calibration sources but achieving and maintaining registration accuracy automatically in the field is a significant challenge. This paper presents an efficient image registration algorithm capable of automatically registering dual waveband image streams upon system start-up and then producing updated transform coefficients during live operation. The algorithm is fully automatic and constrained to ensure reliable operation with minimal or no operator supervision. Robustness to large initial alignment errors is demonstrated using a selection of challenging multimodal image sets. In addition, a novel high performance adaptive image fusion algorithm for maximising fused image quality in the presence of sensor noise is presented.