A framework for modeling and rendering realistic hairstyles

Hairstyles play a critical role in producing realistically looking human characters in computer games and feature films. Synthesizing realistic virtual hairstyles, however, remains an opening problem in computer graphics. First, hair is difficult to model. A realistic model contains tens of thousands of fibers that form complex geometric structure. Second, hair exhibits complex light scattering effects that are difficult to render at interactive speeds. In this thesis, I present a novel hair modeling and rendering framework that effectively resolves both issues. On the modeling front, I present a hybrid image-CAD based solution for creating realistic hair models from imagery data. Unlike existing image-based solutions, our method does not require using special acquisition apparatus. Instead, we directly use hairstyle photos obtained from hair fashion magazines, an abundant and easily accessible resource from the internet. We introduce a new approach called the hair guide surface that is easy and intuitive to model from the input images. We provide a four-view interface to assist the user to interactively design the guide surfaces. From the user defined surfaces, we develop algorithms to automatically generate a dense 3D hair fiber orientation field and then grow hair fibers from the orientation field. On the rendering side, I develop a new approach to simultaneously render various scattering effects including volumetric shadows, transparency, and anti-aliasing on complex hair models. To accurately locate and accumulate the light scattering events along each ray path, we develop a new GPU-based K-Buffer solution. A K-Buffer stores the foremost K fragments along each ray. Compared with existing multi-layer based approaches, we show that our K-Buffer solution can more accurately reproduce the shadow and transparency effects. We further present an anti-aliasing scheme that directly builds upon the K-Buffer. We implement all three effects under a unified framework and integrate it with the modeling module to preview and assess the quality of the model. We show that our framework allows the user to create a wide variety of complex hairstyles with a much lower workload.