A frequency domain approach to fast and accurate image registration

Abstract

Owing to the demanding processing accuracy and speed for image registration in industrial application, an approach to image registration based on Pseudo-Polar Fast Fourier Transform (PPFFT) and Small World Clonal Selection Algorithm (SWCSA) is introduced. We propose a three-step procedure. Firstly, PPFFT is performed to the image. Then, by using the magnitude of PPFFT, a cost function is designed. Lastly, an optimization algorithm called SWCSA is applied to obtain registration parameters. With PPFFT, the complexity of computation is significantly lower and the speed is faster. Because of the optimization algorithm used, the processing accuracy of our method is higher than that of phase correlation algorithm (PCA). Experimental results demonstrate that the speed of the proposed method is two times faster than that of PDFT-PCA and the maximum registration error is less than 0.2°.