Robust Point Correspondence Applied to Two-and Three-Dimensional Image Registration
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
In non-rigid image registration problems, it can be difficult to construct a single cost function that adequately captures concepts of similarity for multiple structures, for example when one structure changes in density while another structure does not. We propose a method that locally switches between cost functions at each iteration of the registration process. This allows more specific similarity criteria to be embedded in the registration process and prevents costs from being applied to structures for which they are inappropriate. We tested our method by registering chest computed tomography (CT) scans containing a healthy lung to scans of the same lung afflicted with acute respiratory distress syndrome (ARDS). We evaluated our method both visually and with the use of landmarks and show improvement over existing methodology.