Non-Photorealistic Rendering in Context: An Observational Study Tobias Isenberg * Petra Neumann * Sheelagh Carpendale * Mario Costa Sousa * Department of Computer Science University of Calgary, Canada Joaquim A. Jorge † Instituto Superior T´ ecnico Lisboa, Portugal Abstract Pen-and-ink line drawing techniques are frequently used to depict form, tone, and texture in artistic, technical, and scientific illustra- tion. In non-photorealistic rendering (NPR), considerable progress has been made towards reproducing traditional pen-and-ink tech- niques for rendering 3D objects. However, formal evaluation and validation of these NPR images remain an important open research problem. In this paper we present an observational study with three groups of users to examine their understanding and assessment of hand-drawn pen-and-ink illustrations of objects in comparison with NPR renditions of the same 3D objects. The results show that peo- ple perceive differences between those two types of illustration but that those that look computer-generated are still highly valued as scientific illustrations. CR Categories: I.3.0 [Computer Graphics]: General Keywords: Non-photorealistic rendering (NPR), evaluation of NPR and traditional scientific illustration, observational study, pen- and-ink illustration. 1 Introduction and Motivation After more than two decades of intensive research, non- photorealistic rendering (NPR) is an established and important field within computer graphics [Gooch and Gooch 2001; Strothotte and Schlechtweg 2002]. NPR breaks free from the constraint of (photo-)realism that many other rendering techniques strive for. It generates images and animations that at least in some parts appear to be made by hand [Strothotte and Schlechtweg 2002]. Within NPR, one exciting direction deals with how computers can be used to generate line drawing illustrations, often with the goal to depict scientific subject matter. For achieving “non-photorealism” in these types of images, NPR takes inspiration from a long tradition of artis- tic and illustrative depiction. Over hundreds of years, hand-drawn scientific illustrations have achieved a high level of sophistication. NPR often tries to imitate long established illustration techniques but we strongly feel that the NPR research has reached a point of sophistication at which it is time to halt and investigate where our research stands compared to hand-drawn illustrations. We conducted an observational study to examine how people under- stand and assess both traditionally created hand-drawn illustrations and computer-generated non-photorealistic illustrations. We chose pen-and-ink illustrations as our study domain since the techniques * e-mail: {isenberg | pneumann | sheelagh | mario}@cpsc.ucalgary.ca † e-mail: jorgej@acm.org have reached a mature state, having been a topic of NPR research for more than ten years. Pen-and-ink illustration is inspired by il- lustration techniques frequently found in hand-drawn scientific il- lustrations and provides a relatively cohesive domain. The purpose of this study was to improve our understanding of the differences between hand-drawn and computer-generated images at the current stage of NPR research and open up new or validate current research directions for the NPR community. We studied how people view both types of images, asked about their imaginable contexts of us- age, assessed participant’s likes and dislikes, and asked about di- rections for improvement of images that were found to be less ap- pealing. To some degree this study is also an attempt of a Turing test for pen-and-ink line drawings since we determined which im- ages were described as having a computer-generated feel to them. Our findings reveal that there are still obvious differences between computer-generated and hand-drawn illustrations. We start by discussing related work in Section 2. Then we explain the design of the observational study in Section 3. Afterwards, Sec- tion 4 presents the results and general observations. In the next Sec- tion 5 we discuss and interpret these results. Finally, we conclude with a summary and suggestions for future work in Section 6. 2 Related Work Relatively few papers have been devoted specifically to evaluations of NPR methods, systems, and images. We consider these studies from six categories of evaluation goals and applications as relevant. Communication in architecture and design: The first NPR evalu- ation [Schumann et al. 1996] studied the usability of computer-ge- nerated images with respect to communicative goals during design concept development with 54 architects as participants. These par- ticipants were shown three different images portraying the same architectural object: a CAD plot image, a constant shading image and a NPR image generated by a sketch-renderer developed by the authors [Strothotte et al. 1994]. Their results show that these three image types have different effects on viewers. For example, NPR images appeared to invite more interaction than the other two types. Space perception in immersive environments: The first experi- ments to examine and evaluate space perception in a functional, non- photorealistically rendered immersive environment [Gooch and Willemsen 2002], involved direct walking tasks in a physical hall- way and NPR renditions (silhouettes, boundaries, and creases) of a 3D model of the same hallway, visualized through a head-mounted display. Their study provided important indications of the degree to which NPR images are capable of conveying a veridical sense of spatial organization. Psychology of NPR: Duke et al. [2003] explored the affective qual- ities of images in a series of experiments. Their main conclusion was that understanding of rendered images requires models that go beyond perception to harness the dynamics of semantic processing in the context of specific tasks. Indeed, they showed through ex- perimental evidence that rendering styles can convey meaning and influence judgment in non-trivial ways.