Hu, QiyangZwicker, MatthiasFavaro, PaoloMatthias Hullin and Marc Stamminger and Tino Weinkauf2016-10-102016-10-102016978-3-03868-025-3-https://doi.org/10.2312/vmv.20161340https://diglib.eg.org:443/handle/10.2312/vmv20161340In this paper we introduce silhouette constraints to improve the quality of unconstrained 3D face reconstruction. Previously, state of the art unconstrained 3D face reconstruction techniques relied solely on photometric consistency and matching sparse facial landmarks. We show that constraining the silhouettes of the 3D reconstruction to match silhouettes in the input images can further improve reconstruction quality. Our technique automatically detects silhouettes and iteratively matches silhouette points computed from the current 3D reconstruction with silhouettes in the input images. We demonstrate that our results improve on the previous state of the art in unconstrained 3D face reconstruction and that our additional constraints can easily be included in the iterative reconstruction at little additional cost.10.2312/vmv.2016134037-43