Research

Highly Detailed Face Capture

The multi-view photogrammetric setup at Camera Facialis.

In-vivo measurement of three-dimensional facial morphology is typically done using photogrammetry. A major challenge is the reliable reconstruction of facial shape and texture from photographs with minimum manual intervention. As part our portrait studio Camera Facialis, we are developing methods for the reconstruction of highly-detailed facial morphology from passive multi-view photographs in fully automated fashion.

Posed facial expressions captured in Camera Facialis and automatically reconstructed using our multi-view reconstruction pipeline.

Dense Face Matching

The identification of correspondence is an essential step in morphological analysis. We are particularly interested in approaches that establish dense correspondence over the entire facial surface reliably and with very high accuracy. We are developing automatic methods for face matching by joint alignment of geometric and photometric features. Our methods achieve accurate correspondence between and within subjects up to the level of smallest skin features like spots and pores (see here).

To establish dense correspondence, individual features of face S are parametrized with low distortion and non-linearly matched to a reference R (details can be found here).

Statistical Face Models

From a large number of faces brought into dense correspondence, we are establishing a statistical face model. To this end, we are investigating theoretical foundations as well as practical approaches to statistically analyze even smallest morphological patterns in shape and texture. The statistically determined inter- and intra-individual patterns (sex, ethnicity, age, expression, etc.) are integrated into the face model. It can be used to plausibly synthesize and manipulate faces by adjustment of the various patterns.

The patterns contained in the statistical face model allow synthesis of diverse facets of facial expressions (for details see here).