Octa-Log-Polar Fourier Transform for Image Registration

A novel image registration algorithm based on octa-log-polar Fourier transform (OLPFT) for estimating large translations, rotations, and scalings in images is presented. The OLPFT, which is calculated at points distributed at non-linear increased concentric octagons, approximated log-polar Fourier representations of images accurately. And it can be calculated fast by utilizing the Fourier separability property and fractional FFT. Using the log-polar Fourier representations and cross-power spectrum method, we can estimate the rotations and scalings in images, and obtain translations later.