Aesthetics of Hand-Drawn vs. Computer-Generated Stippling

Recent work in non-photorealistic rendering has produced results comparable to hand-drawn artistic images. Inspiration for such techniques has come from many traditional artistic techniques, such as pen-and-ink, to depict tone, depth, and shape. These techniques can create visually appealing images and increase understanding as is evident in their use in medical textbooks, popular science, etc. However, when computer-generated images are visually compared to similar hand-drawn images, studies have shown that subjects are generally able to determine differences between both images. This seems to indicate that there are different aesthetics associated with computer-generated images and hand-drawn images. This paper discusses the implications of varying aesthetics amongst hand-drawn and computer-generated images, focusing particularly on the application of stippling to provide tone and shape to an image. 1. Illustrating with Stippling From hastily sketched figures on napkins to complex medical illustrations, hand-drawn images have long been used to convey information to people. Often, the information being presented will be condensed to the most important details by the artist, creating a simple, clear, and meaningful image. This refinement can be accomplished by directing attention to relevant features, simplifying complex features, or exposing previously obscured ones. This selective inclusion of detail provides illustrations with levels of expression not found in photographs. Many techniques found in illustrations have been adopted in an area of computer graphics known as nonphotorealistic rendering (NPR). Particular focus has been given to pen-and-ink techniques which attempt to mimic artists’ strokes, textures and tones through the placement of lines and points of varying thickness and density. To better highlight the differences amongst computer-generated images and hand-drawn images, we will focus solely on stippling as a subset of pen-and-ink. In stippling, dots are deliberately placed on a surface of contrasting color to obtain subtle shifts in value. These dots can be varied in size, volume, and arrangement to create the illusion of different texture, tone, and shape, as shown in Figure 1. Stippling describes both an illustration technique and a visual style. When creating a stipple image, the artist begins by placing dots randomly on the page and then gradually fills in areas from each “seed” dot [Hod89] in order to Figure 1: Examples of hand-drawn illustrations that use stippling. Images copyright of William M. Andrews, used with permission. visually describe forms and objects. This is a purely blackand-white technique, meaning that 100% black marks are made on a 100% white ground, or vice-versa. As such, no gray marks are made. However, because the size and density of marks can be varied, the perception of shades of gray can be readily accomplished. Varying tones are created by placing stipples closer together to form dark regions and further apart to form lighter regions [Nic95]. In this case, individual marks are not important; rather, it is the sum total of the marks which create gray tones. In addition, stippling is capable of capturing a very wide dynamic range of tones, from c © The Eurographics Association 2007. R. Maciejewski, T. Isenberg, W. Andrews, D. Ebert, M. Costa Sousa / Aesthetics of Hand-Drawn vs. Computer-Generated Stippling white to black. This range and subtlety allow very fine gradations between distinct tones. Given the robustness of stippling, many computer illustration systems have developed stippling algorithms. Deussen et al. [DHvOS00] renders polygonal models into a continuous tone image and then converts these target images into a stipple representation. Their work suggests using a Poissondisc distribution to simulate the artistic stipple distribution. In contrast to approximating stipples with a Poisson-disc, Secord [Sec02] uses a fast probabilistic method that places small arbitrarily-shaped primitives, including stippling. Lu et al. [LTH∗02] introduced an interactive stipple rendering system. Here, the initial points for each polygon are generated randomly within the polygon and then redistributed to form a Poisson-disc distribution simulating traditional stipple placement techniques. Meruvia Pastor et al. [MPFS03] present an approach where stipple particles are attached to the surface of the model using a point hierarchy to control the stipple shading density, and Sousa et al. [CSFWS03] present a system which approximates stippling by using short serrated ink strokes that are modeled directly over the edge of the mesh. Schlechtweg et al. [SGS05] created a system that uses a multi-agent system called RenderBots to position NPR strokes, including stipples, based on a stack of G-buffers which are generated from a 3D model. 2. Two Different Aesthetics of Stippling While these non-photorealistic stipple techniques are able to capture many aspects of the styles found in hand-made images, visible dissimilarities can be found between them. In a recent observational study comparing hand-drawn with computer-generated pen-and-ink drawings, Isenberg et al. found that participants were generally able to distinguish between both categories [INC∗06]. For stipple images this was due to differences in stipple point density, the use of shading, and artifacts being present. These artifacts included unwanted regularities in the computer-guided dot placement, leading to the formation of lines as opposed to a more random placement of the dots in the hand-drawn images (Figure 2(a)). They also included intentional artifacts in the dot shape of hand-placed dots as opposed to the very regular and rounded computer-generated dots. These differences may not be purely observational, as an analysis of texture properties between hand-drawn and computer-generated images could potentially be used to classify them into separate categories of hand-drawn/computer-generated if the differences are common within a set. However, Isenberg et al.’s study also showed that the differences between hand-drawn and computer-generated images did not necessarily lead to one category being more appreciated or liked than the other. In contrast, participants said that they like both categories of images for different reasons and would use them in different domains. This leads us to conclude that, while NPR techniques still hold potential for improvement, there may be two different types of aesthetics involved, one in hand-drawn and one in computer-generated stippling. 2.1. The Aesthetic of Hand-Drawn Stippling As a visual style, stippling fills an important role in medical, scientific, and technical illustration. With its ability to depict tonal variations, stippling is well-suited for illustrating objects with fine detail, subtle texture, and small changes in shading. A limitation of line illustration techniques in general is that the individual marks must be smaller than the finest detail that needs to be depicted. In stippling, this is rarely a concern. Gradients and soft edges are relatively easy to create by varying the size and density of marks. However, long lines and hard edges are relatively difficult to create using stipples. For particular illustrations, stippling is the preferred choice because other pen-and-ink techniques, such as hatching, may be mistaken for contours in the images. The random placement of stipples in medical illustrations provides for tone and shape, while not creating any undesired directional cues. This is not to say, however, that creating structures in stippling is always undesired. In terms of stippling and the conceptual approaches an artist might take towards creating stippled illustrations, there are several inter-related issues that the artist must address. First, there are the physical characteristics of the marks, including mark size and variability, mark frequency and variability. Second, there is the issue of edge/shape handling: edge recognition and emphasis is crucial as it effects shape/form recognition and depth cues that arise from shapes interacting with one another. Finally, there is the issue of form/object shading which also serves as a means of emphasizing/de-emphasizing forms. Furthermore, stippling is conceptually like half-tone printing, in which a continuous-tone image (photo or artwork) is converted to discrete dots in order to be printed using only black ink on white paper. The size and density of the dots determine the gray value of the tone printed. Where half-tone conversion typically uses a regularly arrayed screen for conversion, traditional stippling is accomplished through handeye coordination of the illustrator. There is enough variation in the placement of hand-made stipples to appear totally random. In a similar fashion, a stochastic (random or probabilistic but with some direction) screen can be applied to make the half-tone conversion, thus giving a more natural, stippled look to an image. 2.2. The Aesthetic of NPR Stippling Creating stipples in non-photorealistic rendering is typically done by placing points explicitly or using small short strokes that approximate stippling. In Figure 2 we can see that both methods are fair approximations of stippling. Typical NPR stipple creation involves choosing a stipple primitive and a c © The Eurographics Association 2007. R. Maciejewski, T. Isenberg, W. Andrews, D. Ebert, M. Costa Sousa / Aesthetics of Hand-Drawn vs. Computer-Generated Stippling (a) Explicit point placement using [Sec02]. (b) Short strokes generated by