One-Dimensional Mappings for Recovering Large Scale Projective Transformations

One-dimensional mappings in the log-polar and in the inverse-polar spaces are proposed. Based on these mappings, a two-step search strategy has been developed to recover the eight parameters of a general projective transformation between two image patches. First, the four affine parameters are recovered using the one-dimensional log-polar mapping. Secondly, the two projective parameters are recovered using the one-dimensional inverse-polar mapping. At each step the recovery is done by mapping only two line pairs. The remaining two translational parameters are determined by applying these two mappings to an exhaustive search. The proposed mapping strategy has successfully been used to recover different two-dimensional transformations between real images.