Multisensor image registration by feature consensus

Abstract This paper presents an approach for registering images obtained using different sensors, viewpoints or lighting conditions. This approach does not require feature correspondence or area correlation. The geometric transformation between the images is reparametrized into a sequence of elementary stages. At each stage, a single transformation parameter is estimated using a feature consensus mechanism wherein the value of the parameter that is maximally consistent with all possible feature pairings is determined. The concepts of parameter observability and separability are introduced to guide the choice of features, attributes and transformation parameters. Experimental results on real data are provided.