Toward image-based facial hair modeling

Abstract

In this paper we present a novel efficient and fully automated technique to synthesize realistic facial hair---such as beards and eyebrows---on 3D head models. The method requires registered texture images of a target model on which hair needs to be generated. In a first stage of our two-step approach a statistical measure for hair density is computed for each pixel of the texture. In addition, other geometric features such as 2D pixel orientations are extracted, which are subsequently used to generate a 3D model of the individual hair strands. Missing or incomplete information is estimated based on statistical models derived from a database of texture images of over 70 individuals. Using the new approach, characteristics of the hair extracted from a given head may be also transferred to another target.