| 摘要 |
A novel curve evolution approach expressed in a level-set (Fig. 1, 1-2) framework is used to achieve image registration. Implementation of this approach is based on the observation that any coordinate transformation between two given images can be achieved by finding an appropriate mapping (Fig. 1, 4) between two image intensity functions. Therefore, evolving level-sets of one image into the level-sets of other is performed. To estimate the deformation between the images, a level-set motion model is derived which yields the optical flow between the images under certain conditions. One equation which may be used to compute optical flow is a nonlinear hyperbolic partial differential equation, which is a novel formulation for solving the optical flow problem in computer vision. A method for performing multimodel image registration involves matching local frequency image representations. This method involves minimizing, over all affine transformations, the expectation of the squared difference between the local frequence representations of source and target images. In cases where fusing the multimodal data requires estimating the non-rigid deformations, a novel and fast PDE-based morphing technique may be used that will estimate non-rigid alignment. |
