Unambiguous Control Point Selection in Automatic Image Registration Using Energy Cross Correlation, Edge Detection

Abstract

This paper is about how to decide unambiguous control points for automatic image registration. The basic algorithm is
depending on cross-correlation similarity measure. But due to geometrically distorted images this similarity measure need some
good decision capability. So this paper explores the integrity of other parameters for correct points. The image is first divided into
sub images and calculates the energy of each sub image. Based on those values it can be sort out from low energy to high energy
sub image. It also considers the similarity of points using edge detection like canny edge detector. Instead of operating on gray
scale image this algorithm tells to operate on edge image which is opted using canny edge detection method. Both reference and
geometrically distorted images now have only high frequency information which is helpful to find control points matching using
energy of sub images between these two images. once control points match correctly it is very easy to perform transformation on
distorted image.