Non-rigid Registration by Diffusion Model with the Demons Algorithm

Non-rigid registration becomes more and more important in biomedical imaging applications. A novel non-rigid registration method based on diffusion model with demons algorithm is proposed in this paper. The moving image is considered as a deformable grid, and it is diffusing through the contours of the objects in itself, by the action of effectors, called demons, situated in these interfaces. In order to speed up the convergence of the iteration and enhance the robustness of the algorithm, the moving image is smoothed to increase the number of the ‘demons’. The deformation field has been normalized at each iteration to ensure the stability. Our algorithm is compared with the original ‘demons’ and other developed algorithms on synthetic images and real medical images, the result shows that our algorithm has obvious speed improvement and high tolerance of large deformation and noise.