Hair Modeling and Synthesis

Hong Chen 1 and  Song-Chun Zhu1,2
1University of California, Los Angeles;  2Lotus Hill Institute    

1. The Problem

Human hair is a very complex visual pattern where hundreds of thousands of hairs are grouped into strands and wisps in diverse hair styles. Modeling hair appearance is an important and challenging problem in graphics, digital human animation, and on-photorealistic rendering.

2. Solutions

We present a generative sketch model for human hair analysis and synthesis. We treat hair images as 2D piece wisely smooth vector (flow) fields, and thus our representation is view-based in contrast to the physically based 3D hair models in graphics. The generative model has three levels. The bottom level is the high frequency band of the hair image. The middle level is a piece wisely smooth vector field for the hair orientation, gradient strength, and growth directions. The top level is an attribute sketch graph for representing the discontinuities in the vector field. Besides the three level representations, we model the shading effects, i.e. the low-frequency band of the hair image, by a linear superposition of some Gaussian image bases, and we encode the hair color by a color map.

3. Hair Model and Synthesis

The generative model for both analysis and synthesis is illustrated in the Fig below. There are three factors contributing to the appearance of hair: (i) hair color, (ii) shading effects, and (iii) texture. Therefore a hair image is first decomposed into these three components. Overview of our model and algorithm which consists of three modules: (i) analysis, (ii) synthesis, reconstruction or rendering, and (iii) an optional editing interface. For an input image , we decouple it into a gray image  and a color channel  represented by a color map. The gray image is decomposed into a texture part  and shading  by a low-pass Gaussian filter. From we compute the vector field V and the sketch with its direction. The synthesis goes from the sketch S to the vector field  and to the hair image. The latter is combined with the shading  and color  to produce the final result .

There is an example of hair model and inference. (a) is an input color image , (b) is the computed sketch S with directions dS. (c) is the sketchable vector field  generated from (S; dS). (d) is the overall vector field V after filling-in non-sketchable part. (e) is the high frequency hair texture image  generated from the vector field. (f) is the shading and lighting image. (g) is the synthesized color image n after adding the shading and color. We render an artistic sketch  in (h).

4. Some Synthesis Results


5. Publications

H. Chen and S.C. Zhu, "A Generative Sketch Model for Human Hair Analysis and Synthesis", IEEE Trans. on Pattern Analysis and Machine Intelligence, vol.28, no.7, 1025-1040, July, 2006. [pdf]