Abstract

Automated 3D reconstruction of faces from images ischallenging if the image material is difficult in terms ofpose, lighting, occlusions and facial expressions, and if theinitial 2D feature positions are inaccurate or unreliable. Wepropose a method that reconstructs individual 3D shapesfrom multiple single images of one person, judges theirquality and then combines the best of all results. This isdone separately for different regions of the face. The coreelement of this algorithm and the focus of our paper is aquality measure that judges a reconstruction without infor-mation about the true shape. We evaluate different qual-ity measures, develop a method for combining results, andpresent a complete processing pipeline for automated re-construction.