A number of works on nonphotorealistic rendering were introduced by the David Salesin’s group at the University of Washington. Instead of defining lines individually with the path-and-stroke metaphor, so-called “stroke textures” are introduced. Here, a texture consisting of strokes is generated for various gray scales and viewing angles. This texture can be applied for automatic rendering [232, 233] as well as in the context of interactive drawing programs [182, 183, 184].
In  a tree sketch is modeled with this method. Hereby the directions of the foliage and trunk silhouette are modified by a given vector field, and by an additional gray value image. Figure 11.2 shows the tree sketch as well as the applied data. The user defines the grey value image, for which a photograph can be used. Additional inputs are the stroke texture itself as well as the vector field for their orientation. The angles are defined by colors of a corresponding texture.
For the production of the image, the so-called difference-image algorithm is used. The algorithm places textures or strokes in the resulting image until the differences in the grayscale values between the given and the resulting image are sufficiently small. Here a simple optimization algorithm is used for the filling that adds new strokes at the places with the largest differences. In this sense the method is a half-toning technique with a greater freedom in the choice of the line objects that are used for drawing. In a similar way it is possible to illustrate other objects.
The results are of a high aesthetic quality; however, this procedure is not suitable for the automatic rendering of landscapes. An important aspect of this work is the use of different pictures or buffers for the production of the strokes in the illustrations; this is an extension of the Saito and Takahashi approach, and will be used in the following at several places.
Plant Pictures in the Style of Children’s Books
Kowalski et al.  introduce a method to illustrate plant scenes in the style of two well-known authors of children’s books. In contrast to the already – mentioned procedures of Salisbury et al. , here we deal with an automatic method that also uses a 3D model as its basis.
For rendering the image, the authors apply a multilevel algorithm, which in the first step illustrates the scene conventionally. The grayscale values of the created image are the starting point for the placement of so-called “graftal textures”, which are positioned at those places that appear dark in the initial image. Here the authors utilize the difference-image algorithm introduced by Salisbury et al.
A graftal texture contains strokes or other shaping elements, but also an algorithm for the production of graphical objects can be called up. A special ad-
vantage here is that the sizes of the objects can be selected to reveal the correct perspective distortion or to be constant.
For the first image of a sequence, the objects are chosen according to this scheme. For the following images, on one hand the system tries to position the objects due to the new positions of the corresponding elements in the scene, and on the other hand to achieve a possibly high coherence between the frames. Once the surfaces of an object are retrieved and the old positions of the graftal textures, the authors are then able to choose the new positions similarly. However, here they have to consider old graftals that can be left out and new graftals that possibly appear. However, the so-obtained image sequences show that the goal of spatial and temporal coherence is only reached partially using this method.
In Fig. 11.3 the method is sketched. Area (a) shows the initial scene, area (b) the respective nonphotorealistic result. particularly the foliage of area (c) is a good rendition in the sense of illustrations. A disadvantage of the method is that the initial models must represent the foliage of a tree through large balls that have to be specially produced, and thereby the use of realistic tree models is excluded.
A generally applicable procedure should use such surface models, of which there are many. Also, the application areas such as architecture, landscaping, and biology create more requirements: it should be possible to vary the level of abstraction and to execute the transition from the illustrative rendering to a realistic representation. Additionally, the already-mentioned spatial and temporal coherence between the individual pictures of a sequence should have been established, so that the illustrations generated can be shown interactively or as animations.
In Sect. 11.3 we introduce a solution for this problem. It is based on the analysis of traditional drawing methods from various illustrations. Thus in the following we will examine what kinds of effects or styles are actually used in hand-drawn illustrations, and also up to what degree it is possible to formalize and transfer the findings into an executable algorithm.