Robust image registration using phase congruency

Image registration plays an important role in many image processing applications. This paper proposed a robust image registration method that can estimate motion parameters including translation and rotation from a group of different-exposed images. Brightness change and photometric distortion, i.e., under-exposed or over-exposed, between images will result in inaccuracy of the estimation. To solve this problem, we propose to firstly transform raw image data into phase congruency that is invariant to the changes in image brightness or contrast and then apply phase cross correlation in the Fourier transform domain to estimate motion parameters. Experiment results show that the proposed algorithm is efficient and robust against photometric distortion compared with other phase correlation based registration algorithms.