Features based image registration using cross correlation and Radon transform

Abstract

In many research fields, like Medical image analysis, pattern recognition, computer vision and remote sensed data processing, it is required to align the images. This article describes and compares two different methods to register images. In the first method, the Fourier Merlin transform based on the phase correlation of the two images in the Log-polar domain. However, it suffers from non-uniform sampling which makes it not suitable for applications causing losses in image information which will decrease the registration accuracy. The second method is to use the Radon transform proprieties to extract the rotation angle. Experiments show that the proposed method can detect the angle of rotation efficiently for different types of images without the SAR complex images.