3D Face Modeling (Images)

We reconstruct a distinctive 3D face model, given a collection of facial images, for challenging unconstrained face recognition problem. The key idea is to build a 3D shape on each image and combine all reconstructions into the final shape with a pooling method considering the quality of each input image. We first build 3D model using a robust variation of the state of the art 3D Morphable Model (3DMM) fitting algorithm on each input image. We then use a simple but effective fusion method to attain a single 3D output from multiple sources, considering the qualities and confidences.

Technical Report: Reconstruct distinctive 3D face model from multiple images [PDF]