Realistic, Hardware-accelerated Shading and LightingWolfgang Heidrich∗Hans-Peter SeidelMax-Planck-Institute for Computer ScienceAbstractWith fast 3D graphics becoming more and more available even onlow end platforms, the focus in hardware-accelerated rendering isbeginning to shift towards higher quality rendering and additionalfunctionality instead of simply higher performance implementa-tions based on the traditional graphics pipeline.In this paper we present techniques for realistic shading andlighting using computer graphics hardware. In particular, we dis-cuss multipass methods for high quality local illumination usingphysically-based reflection models, as well as techniques for the in-teractive visualization of non-diffuse global illumination solutions.These results are then combined with normal mapping for increas-ing the visual complexity of rendered images.Although the techniques presented in this paper work at interac-tive frame rates on contemporary graphics hardware, we also dis-cuss some modifications of the rendering pipeline that help to fur-ther improve both performance and quality of the proposed meth-ods.CR Categories: I.3.1 [Computer Graphics]: HardwareArchitecture—Graphics processors; I.3.3 [Computer Graphics]:Picture/Image Generation—Bitmap and frame buffer operations;I.3.6 [Computer Graphics]: Methodology and Techniques—Standards I.3.7 [Computer Graphics]: Three-Dimensional Graph-ics and Realism—Color, Shading, Shadowing and TextureKeywords: reflectance functions, illumination effects, shading,texture mapping, rendering hardware, frame buffer techniques1 IntroductionUntil recently, the major concern in the development of new graph-ics hardware has been to increase the performance of the tradi-tional rendering pipeline. Today, graphics accelerators with a per-formance of several million textured, lit triangles per second arewithin reach even for the low end. As a consequence, we see thatthe emphasis is beginning to shift away from higher performancetowards higher quality renderings and an increased feature set thatmakes graphics hardware applicable to more demanding applica-tions.Despite this development, most current graphics hardware stillonly uses a local Phong illumination model, which was shown to∗Computer Graphics Group, Im Stadtwald, 66123 Saarbr¨ucken, Ger-many, {heidrich,hpseidel}@mpi-sb.mpg.debe inappropriate a long time ago [4]. Moreover, techniques for vi-sualizing non-diffuse global illumination are not widely applied,although several researchers have worked on such methods.In this paper, we present a class of techniques to improve thequality of shading and lighting in hardware-accelerated computergraphics. We start with algorithms for high-quality local illumina-tion using alternative lighting models such as the one by Torranceand Sparrow [39]. This approach is based on an analytic factoriza-tion of the respective model into bivariate terms that can be rep-resented as texture maps. We then discuss methods for visualiz-ing non-diffuse global illumination solutions based on environmentmaps. We introduce both a Fresnel term for simulating reflectionsin non-metallic objects, as well as a pre-filtering method for envi-ronment maps. To this end, we also review an alternative parameter-ization for environment maps that we recently introduced [20], andthat allows us to use one map for all viewing positions and direc-tions. These techniques are finally combined with normal mappingto increase the visual complexity of the scene.Although the techniques presented here produce renderings atinteractive frame rates on contemporary graphics hardware, a moredirect support by future hardware could be achieved through somemodifications of the rendering pipeline. These will be outlined atthe end of this paper.2 Previous WorkMost of today’s graphics hardware uses either the Phong [7], orthe Blinn-Phong [4] illumination model. Many other, physicallymore plausible models have been proposed, but have so far onlybeen used in software rendering systems or by hardware that hasprogrammable shaders, such as [30]. The most important of thesemodels are the ones by Torrance and Sparrow [39] and Cook andTorrance [11]. The Torrance-Sparrow model uses a Gaussian microfacet distribution function and a geometric attenuation factor basedon the assumption of v-shaped grooves on the surface. Other dis-tribution functions have been proposed, for example, by Beckmannand Spizzichino [3], while Smith [36] presented a more accurategeometry term under the assumption of a Gaussian facet distribu-tion. He et al. [18] proposed the HTSG model based directly on theKirchhoff theory of electromagnetic fields. This model is capableof simulating even more physical effects, although at significantlyincreased computational cost.In addition to the isotropic models listed above, anisotropic mod-els have also been proposed. Banks [1] described a very simplemodel based on results from the illumination of lines in 3-space,while Cabral et al. [8] and Poulin and Fournier [32] use simula-tions of different kinds of micro geometry. Ward [42] modified theTorrance-Sparrow model by using an anisotropic micro facet distri-bution function.Several different methods for interactively visualizing non-diffuse global illumination have been suggested in the literature.Environment maps as a means of computing a mirror term wereproposed by Blinn [6], while the spherical parameterization used bymost graphics hardware today was presented by Haeberli and Se-gal [16]. Diefenbach [13] demonstrated a multipass method for ren-dering mirror and glossy reflections on planar surfaces. Ofek andRappoport [29] interactively compute mirror reflections off curvedreflectors by warping the reflected geometry. Miller et al. [28] pro-pose an approach where the glossy reflection on a surface is storedin a compressed light field, and Walter et al. [41] place virtual lightsin the scene to simulate glossy reflections. St¨urzlinger and Bas-tos [38] employ multipass rendering to visualize the result of a pho-ton tracing algorithm. Finally, in some work developed in parallelto ours, Bastos et al. [2] use textures and filtering techniques to ren-der reflections in planar objects based on physically correct reflec-tion models, and Cabral et al. [9] propose an environment mappingtechnique for glossy reflections not unlike ours.Bump maps, originally introduced by Blinn [5], have recentlyfound their way into hardware-accelerated
View Full Document
Unlocking...