Registration of Surfaces to 3D Images Using Rigid Body Surfaces

We propose a constrained active surface model, called the rigid body surface, that deforms via rigid body transformation. In each iteration, the surface is deformed first via partial differential equations and then the least-squares estimate of the rigid body transformation between the deformed surface and predefined surface is calculated. The rigid body transformation estimate is based upon the singular value decomposition. Experimental results on surface-to-3D image and surface-to-surface registration illustrate the insensitivity of the rigid body surface to noise and initialization. In addition to registration, rigid body surfaces can be employed for image segmentation and tracking when the target shape is predefined.