Testing the Robustness of JAYA Optimization on 3D Surface Alignment of Range Images: A Revised Computational Study

Resumen

The further development of image registration (IR) as automated image alignment techniques is a well-known concern in the field of computer vision (CV). These techniques have been implemented in many real-world scenarios: from remote sensing to medical imaging to artificial vision and computer-aided design. There is great interest in applying original optimization algorithms to overcome the challenges associated with early IR methods (e.g. the ICP algorithm). On the other hand, algorithms rooted in evolutionary theories in nature-inspired computational models such as Evolutionary Computing (EC), have gained in importance over the past two decades. In addition, other algorithms with wide applicability fall into this category of methods, e.g. metaheuristics, swarmming, etc. Most of these methods have been widely adopted to address the IR problem, and are used as reliable alternatives for optimization goals. The aim of this paper is to address the following two main research challenges: (i) Bridging the gap in the revision of solutions proposed in recent years from the latter optimization model and from those ones from a new model based on deep learning (DL); and ii) introducing a new non-metaphor-based IR approach, called JAYA, to solve specific problems of aligning 3D surfaces of range images, also known as range image registration (RIR). In fact, as far as is known, this is the first time JAYA has been suggested for the above RIR problem. In particular, a new RIR method using the JAYA algorithm have been introduced and its performance has been accordingly compared against a wide set of methods from the SoTA. More than a dozen Softcomputing-based RIR methods have been included in the experimentation, making it the largest comparative study ever carried out in this category. In particular, range image datasets belonging to the SAMPL repository have been used, which has been widely adopted by many authors in the SoTA.