关键词： 图像配准, 全变分, 保留不连续性, 非线性多重网格, 滞后扩散不动点迭代, 逐次超松弛迭代

Abstract: An improved total variational (TV) registration  model with both global smooth and discontinuous-preserving displacements is established by introducing hypersurface function as the kernel function of image registration regularization term. In the nonlinear multi-grid (NMG) method, a new smoothing method is constructed by using the interaction of delayed diffusion fixed point iteration and successive over relaxation iteration. And by interpolation of the errors on the coarse grid back to the fine grid using tomographic techniques during confinement, a fast and effective nonlinear multi-grid algorithm for solving improved TV model is designed. The experimental results show that the optimized NMG algorithm has higher registration accuracy and faster convergence than the NMG method. Improved TV model registration has less error, less time-consuming and better registration performance than TV model registration.

Key words: image registration, total variation, discontinuity-preserving, nonlinear multi-grid, delayed diffusion fixed point iteration, successive over relaxation iteration