Hough transform can give you features or keypoints (lines, circles, ellipses)..
you will need a similarity metric, a matching algorithm and an aggregator over keypoints (such as an affine transform fit with RANSAC) to register the 2 images.
Thank you for the answer but, the images which I want to register do not contain any line ,curve or any other shapes. Images are orientation fields of two different Palmprints and I want to register them roughly. In a paper, I saw that the author had referred to Generalized Hough Transform as the registration method.
Registration is a technique to align images to a plane with center (0,0) being the optical axis of the camera center (imagine 4 quadrants). Now, here what is done is we take the pixel values(unaligned) and transform it to fit this plane.
So for planar transformation (Cartesian coordinates - x,y) we have to build a R-table that transforms it to (x',y') that has the (close to) precise gradient magnitude of the object.
Note: R-table is for identifiable object(s) in the image
Bk- Feature Extraction and Image Processing - M Nixon & Alberto S. Aguado