Self-adaptive evolutionary image registration using differential evolution and artificial immune systems

Image registration is present in many computer vision and computer graphics real-world applications. Specifically, it plays a crucial role within the 3D digital model acquisition pipeline, in which the iterative closest point (ICP) algorithm is considered the de facto standard for pair-wise alignment of range images. Nevertheless, the success of ICP depends on several assumptions. A new family of registration techniques have been recently proposed based on evolutionary computation paradigm to solve the common ICP problems.Unlike previous contributions, we propose a novel self-adaptive evolutionary image registration algorithm able to search for the values of both the control and the problem solving parameters to achieve accurate alignments, simultaneously. It combines two different population-based optimization approaches that are concerned with the proper optimization of the control parameters and the image alignments, respectively. The performance of our proposal is compared with several state-of-the-art image registration methods.

► IR aims to find a geometric transformation between two or more images. ► In the last decade, the application of evolutionary algorithms to IR has caused an outstanding interest. ► We propose a new self-adaptive evolutionary approach to tackle IR.