Automated Registration of 3D Faces using Dense Surface Models Tim J. Hutton 1 , Bernard F. Buxton 2 and Peter Hammond 1 1 Biomedical Informatics Unit, Eastman Dental Institute, University College London, 256 Gray’s Inn Road, London WC1X 8LD 2 Department of Computer Science, University College London, Gower Street, London WC1E 6BT T.Hutton@eastman.ucl.ac.uk Abstract Dense surface models can be used to register unseen surfaces, using an al- gorithm which is a hybrid of iterative closest-point (ICP) and active shape model (ASM) fitting. In this paper we give details of this procedure and show how it can be improved by sequentially extending the transform group over which it operates. We also evaluate it for robustness to the position of the target and to shape variation across a set of unseen examples. The fit was successful on all 21 examples in our test set, with an average RMS error of 3.0mm. An initial comparison of 3 people landmarking the same scans suggests that this is within the normal landmark reproducibility range for 3D face scans. 1 Introduction The technology for non-invasively acquiring three-dimensional surface scans of biolog- ical subjects, especially of the human face, is becoming widespread. Such scans are medically useful as a means of capturing a detailed record of an individual’s face at a moment in time, for example to audit the outcome of surgery or to analyse facial growth associated with different genetic conditions. To analyse these surfaces usefully they must be registered. However, their manual annotation with landmarks can be time-consuming and subject to error. A technique that could accurately and robustly register surface scans of the human face would find ready applications in medical diagnosis, security, human motion tracking and animation. Many techniques exist for registering surfaces automatically using features such as local curvature [1, 11, 21, 15, 20, 12] but they do not use a model of the range of shape variation expected in the target surface. Where the shape of the target is known to vary significantly, especially for biological objects such as the human face, it seems inevitable that such methods will register unseen surfaces less accurately. Point distribution models (PDMs) have proved to be a useful way both of capturing shape variation and of using it to regularise the search for matching structures in images