COMPARISON OF RIGID REGISTRATION WITH DIFFERENT OPTIMISATION TECHNIQUES

Abstract

The process or technique of matching the appearance of two or more images by determining an alignment between them is known as image registration. This is basically aligns two images geometrically. In this study, two-dimensional image registration is presented using the rigid group. This group is a finite dimensional group under composition; it is four-dimensional in this particular case. The dimensions of the rigid group include scaling, rotation, and translations along the axes. This paper presents methods that use a discretized objective function to construct rigid transformations. Based on SSD (sum of the squares of the distances between the pixel intensities), this objective function calculates the difference between the images. In this paper we compare two image registration optimization algorithms and applied them to six different image registration examples. One is the coarse search, and the other is the gradient descent, implemented using MATLAB’S least-squares optimisation (lsqnonlin). The coarse search algorithm is used to explore the transformation domain and provide an initial estimate, which is then refined using the gradient descent method. The proposed algorithms is implemented on a variety of images, mostly our own captured images. The numerical examples demonstrate that the combined use of coarse search and gradient descent improves registration accuracy in several cases, although both methods may still be affected by local minima in certain situations.