Pattern Recognition and Image Analysis
Hi-index | 0.00 |
This paper describes a new approach that registers two images based on a physically deformable model. No prior knowledge about the types of geometric transformations is required for registration. Instead, our approach simulates the manual registration process and models the registration as the deformation of an elastic material governed by physical laws. Before registration, the contours of the images are extracted. The registration consists of two steps. The physically deformable model is applied first to deform one contour, called the start contour, to match the other, the goal contour. The result is point correspondences between the contours. Then an intensity transformation is estimated from the matching points to warp one image towards the other.