Anatomically Plausible Surface Alignment and Reconstruction

Abstract

With the increasing clinical use of 3D surface scanners, there is a need for accurate and reliable algorithms that can produce anatomically plausible surfaces. In this paper, a combined method for surface alignment and reconstruction is proposed. It is based on an implicit surface representation combined with a Markov Random Field regularisation method. Conceptually, the method maintains an implicit ideal description of the sought surface. This implicit surface is iteratively updated by realigning the input point sets and Markov Random Field regularisation. The regularisation is based on a prior energy that has earlier proved to be particularly well suited for human surface scans. The method has been tested on full cranial scans of ten test subjects and on several scans of the outer human ear.