Registration of multi-focus images by combining global and local transformation models

Abstract

Image registration is a necessary pre-requisite for multi-focus image fusion because before fusion the constituent images must be positioned properly with respect to a common coordinate system. A two-step registration method is proposed in this work. In first step, a global translation is determined by maximizing the mutual information of the source and the reference images and then it is applied on the source image. In second step, a block-wise local scaling is applied on the translated source image. The global and local transformations constitute a hybrid technique which is iterated to obtain the optimal result. Proposed method is easy to implement, requires no manual intervention and gives good results. Results obtained by the method on different sets of multi-focus images are provided with. Performance of the system is also evaluated and is compared with two different methods.