High resolution estimation of planar rotations based on Fourier transform and radial projections

This work presents an original approach for estimating planar rotations between image pairs which resorts to the Fourier transform, in order to avoid the use of features, and to radial projections, in order to reduce the original 2-D estimation problem into a 1-D problem. Our procedure is of coarse-to-fine type and consists of two passes. A first rough estimate of the rotational angle is computed by means of a phase correlation type algorithm applied to the radial projections of the Fourier transforms magnitudes of the two images. This estimate is then refined by computing, at higher resolution, the angular shift between the highest maxima of the radial projections after back rotating the second image according to the first rough estimate of the rotational angle. The accuracy reached by the proposed algorithm is a few hundredths of degree over a range of 360° Experimental evidence of this accuracy is reported