基于拟共形理论的分数阶多尺度微分同胚图像配准

Multi-Scale Approach for Diffeomorphic Image Registration with Fractional-Order Regularization Based on Quasiconformal Theory

图像配准领域存在着两大挑战:(1)网格重叠现象;(2)贪婪配准问题不适定.针对这两大挑战,该文提出了一个基于拟共形理论的多尺度分数阶微分同胚图像配准模型,该模型在无网格重叠及先验正则项的前提下,得到了相似性度量泛函的一个光滑极小值点.此外,该文证明了所提模型解的存在性及多尺度方法的收敛性,并通过数值实验验证了所提算法能有效避免网格重叠并得到较好的配准结果.

Two significant challenges relating to image registration include the issue of mesh folding and the unresolved problem of greedy registration.To tackle these challenges,a multi-scale approach for diffeomorphic image registration model with fractional-order regularization based on quasiconformal theory is proposed in this paper.It employs fractional-order differential to achieve a smooth energy functional minima without mesh folding and a priori regular terms.Furthermore,the existence of solution for the proposed model and the convergence of the multi-scale approach are proved.And numerical tests are performed to demonstrate that the proposed algorithm effectively eliminates mesh folding and generates superior registration results.