This paper proposes a novel automated method for global alignment of three dimensional MR images. The matching algorithm employed is closely related to a common constraint based tree searching algorithm [1], but uses a novel multi-resolution encoding of the search space to improve the search time and permit searching of curved surfaces. The algorithm usees the shape index defined by Koenderink [2] which provides the very useful property of invariance to uniform scale. The surfaces of the objects are extracted from the MR images automatically using a 3D deformable model [3]. An intelligent mechanism is used for selecting unusual surface features that are common to both objects.
展开▼