Fast Occlusion Sweeping

While realistic illumination significantly improves the visual quality and perception of rendered images, it is often very expensive to compute. In this paper, we propose a new algorithm for embedding a global ambient occlusion computation within the fast sweeping algorithm while determining isosurfaces. With this method we can approximate ambient occlusion for rendering volumetric data with minimal additional cost over fast sweeping. We compare visualizations rendered with our algorithm to visualizations computed with only local shading, and with a ambient occlusion calculation using Monte Carlo sampling method. We also show how this method can be used for approximating low frequency shadows and subsurface scattering. Realistic illumination techniques used in digitally synthesized images are known to greatly enhance the perception of shape. This is as true for renderings of volume data as it is for geometric models. For example, Qiu et al. [1] used full global illumination techniques to improve visualizations of volumetric data, and Stewart [2] shows how computation of local ambient occlusion enhances the perception of grooves in a brain CT scanned dataset. Tarini et al. [3] observed that perception of depth for large molecules was significantly improved with the use of ambient occlusion as compared to standard direct shading methods even when coupled with other techniques such as depth cueing and shadowing. Recently, a carefully designed experimental study by Weigle and Banks [4] definitively demonstrated that physically-based global illumination is a powerful adjunct to perspective projection in aiding human subjects to understand spatial relationships in a complex volume rendered scene. Despite the strong evidence for its efficacy in conveying spatial information in visualization, the use of global illumination is rare in practical visualization systems. This is most likely due to the high overhead of existing global illumination rendering algorithms. In this paper, we provide a new solution for ambient occlusion computation that is significantly faster than existing techniques. The method integrates well with a volumetric ray marching algorithm implemented on the GPU. While not a full global illumination solution, ambient occlusion provides a more realistic illumination model than does local illumination, and permits the use of realistic light sources, like skylights. For accelerating our ray marching algorithm, we build a volumetric signed distance field using the fast sweeping method, and we embed our ambient occlusion approximation ? e-mail: [email protected] ?? e-mail: [email protected] ??? e-mail: [email protected] 2 Mayank Singh, Cem Yuksel, and Donald House directly into the sweeping algorithm. Thus, our algorithm can produce an ambient occlusion estimate with only a minor computational overhead. We are also able to use our approach to approximate low-frequency shadows due to direct illumination from certain angles, and to approximate subsurface scattering effects.