Non-photorealistic Rendering

Introduction

Through the careful use of a brush, a skilled watercolor artist is able to distribute semi-transparent pigments on a sheet of paper in a manner that creates a meaningful image. An artist can paint surfaces in a way that conveys the shape and texture of an object that might not be visible in reality. In addition, through the careful selecting and mixing of a limited palette of paints, an artist can present a sense of color harmony while at the same time adding a distinct mood to a scene. In this way, an artist can create a painting that may not be true to reality, but is rich with information. Our work uses ideas from watercolor painting to communicate information about objects in a scene that might not be visible with more photo-realistic rendering methods. The goal of our work is not to produce realistic watercolor pictures, but rather it is to use watercolor techniques that an artist might employ to convey perceptual clues that could help illustrate shape and spatial relationships. Our work does not attempt to use painterly rendering for its own sake, but rather as a means to present more information.

1. An iterative nonphotorealistic rendering process using a watercolor approach.

Our method builds up a scene by iteratively compositing layers of watercolor paint as shown in Figure 1. Each iteration requires the computation of a texture for that layer, and the determination of how the thickness of the paint in that layer should vary across the image. The texture for each layer is computed using Line Integral Convolution (LIC) [2] along one of the principal curvature directions. Then the layer thickness is determined using a modified Phong shading model. Since lighting is performed through the process of mixing pigments, a subtractive color model is used. This requires an inverted lighting model that indicates illumination not through the adding of color where an object is lit, but rather through the adding of paint based on an object's darkness.

Related Work

Several authors have described the use of principal curvature directions for the perceptually meaningful nonphotorealistic rendering of surfaces. Interrante [9] describes how LIC along the principal curvature directions can be used to create brush stroke textures. Her work deals with the creation of individual brush strokes and describes issues involved in selecting their density, thickness, and color to produce images that convey information for scientific visualization applications. In a following work by Girshik and Interrante [5], lines are rendered on a surface by tracing along one of the principal curvature directions. Hertzmann and Zorin [8] describe an algorithm for the rendering of smooth surfaces in a line-art style. They create hatching patterns along the smooth curvature direction fields to communicate shape information. They also describe how C2 subdivision surfaces and their principal curvature directions can be computed. Our work creates painterly textures for subdivision surfaces. Our algorithm first calculates raw principal curvature direction vectors using the method described by Hertzmann and Zorin [8]. These directions are then assigned a ...