A Novel Consistency Regularizer for Meshless Nonrigid Image Registration

In nonrigid motion analysis, deformation fields are often modeled using splines defined on a control-point grid. Inspired by recent development of meshfree methods, we propose a novel motion model that does not use control-point grids, nor use explicit node connections. We also propose a regularizer for the deformation field and the minimization algorithm. The method has promising features such as the handling of irregular regions, adaptive accuracy, the multi-scale modeling, and the potential for integrating physical properties into the registration process. Promising results were obtained on both synthetic and real imagery.