Image Deformation Estimation via Multiobjective Optimization

The free-form deformation model can represent a wide range of non-rigid deformations by manipulating control point lattice over the image. However, due to large number parameters, it is challenging fit directly deformed image for estimation because complexity fitness landscape. In this paper, we cast registration task as multi-objective optimization problem (MOP) according fact that regions affected each overlap with other. Specifically, partitioning template into several and measuring similarity region independently, multiple objectives are built thus be realized solving MOP off-the-shelf evolutionary algorithms (MOEAs). addition, coarse-to-fine strategy pyramid combined mesh subdivision. optimized candidate solutions current level inherited next level, which increases ability deal deformation. Also, post-processing procedure proposed generate single output utilizing Pareto optimal solutions. Comparative experiments on both synthetic real-world images show effectiveness usefulness our method.