Initial Matching of Multiple-View Images by Affine Approximation of Relative Distortions

摘要

To match multiple views of a 3D scene, their relative geometric distortions have to be taken into account. We assume the disortions can be approximated by affine transformations. Images are matched by combining an exhaustive and directed unconstrained Hooke-Jeeves search for affine parameters, image pyramids being used to accelerate the search. The parameters found for several matches are statistically processed to relatively orient the images. Experiments with the RADIUS multiple-view images show a feasibility of this approach.