Towards Physically-Sound Registration Using Object-Specific Properties for Regularization

We study a landmark-based image registration technique that uses elastic model and boundary mapping. Emphasis is given to the reconstruction of heterogeneous material properties and the use of recovered object-specific information to facilitate the registration computation. Tikhonov functional is used to estimate elastic property from landmarks with measurement uncertainty. A preliminary study using a 2D synthetic object indicates that incorporation of actual material properties in the registration method can improve registration accuracy and reduce computational cost.