Introduction

To achieve the range of effects in use today, modelers and animators often engage artists to create custom textures. Currently we lack tools which would allow the non-artist to emulate the nonlinear shading found in works of art. The technical goal of this paper is to create methods for capturing and interactively editing shading models. The artistic goal is to apply to geometric objects and scenes the subtle shading effects artists use to infuse works of art with vitality and beauty. This paper presents algorithms for mapping shading from works of art to geometric models, allowing the user to interactively render scenes based on the style of a reference art work.

  

Figure 1: An artist first develops a shading study on a sphere (left) and then adapts this study to a complex object (right). This process allows the artist to first concentrate on color/shading alone, and then to apply the same shading to a complex object in a way that guarantees some global consistency in the final image. Image courtesy of McGraw-Hill Book Company [11].

When an artist draws or paints an object, they often start with a shading study on the sphere [9]. This is shown for a pencil drawing in Figure 1, where an artist uses a reference drawing of a sphere to draw a head with similar shading. Using the sphere in this way ensures that when the complex object is rendered it will have global consistency in its shading. This paradigm of using a reference sphere for shading can be applied in computer graphics. Given an image of a shaded sphere, we can infer how the sphere normals correspond to colors, and shade a model based on local surface orientation.

While transferring a shading model from an image of a sphere to a complex 3D model is straightforward, it would be much more useful to take an image of a complex object and transfer its shading to another complex object. In this paper we use the shading study on a sphere as an intermediate step to help the user ``capture'' a shading model from an image of a complex object. This essentially inverts the traditional process illustrated in Figure 1.

In Section 2 we review related work and introduce the ``lit sphere'' model. Section 3 discusses implementation methods for applying the model to non-photorealistic rendering, and describes the user interface. We present results and some applications for our approach in Section 4. We suggest areas for further investigation and conclude the paper in Sections 5 and  6.