12/2 Registration and predictive modeling of human faces

		Manipulation of facial attributes

A core requirement throughout most procedures for planning and monitoring facial surgery is a computer-aided comparison of head data. Visualisation of differences between data of the same person recorded before and after a surgery are required to monitor heeling and the result of the intervention. Or it might be required to compare the data of a patient with a reference data set to measure the deviation of the patient data to the norm.

Here in this project we approach two of the key limitations for a fully automated data comparison. These are the problem of unreliable non-rigid data registration and second not well-defined reference data sets. First, we start to develop a reference data set representing the statistical variability of face and head appearances within the population. A database with registered scans of individuals enables us to build statistical models of the variations in human head structure. We plan to perform a systematic data collection covering a broad variety of people within the population. Using the acquired data we derive statistical models for various groups of different age, gender or ethnic affiliation. Our first experiments will investigate a parameterisation the face variation according to age of a certain individual. This requires novel methods to separate variations between individuals from variations according to different age. Second, the statistical model on the appearance of faces will be used to develop novel registration methods for a better quantitative data comparison.