Architectures for Realistic Volume Imaging

In this work, we propose architectures that enable increased realism in volume rendering. The real world is a complex environment with complex illumination. We feel that a volumetric representation is ideal for many applications and that compelling realism is possible using advanced rendering algorithms. This work attacks the problem of increasing realism on several fronts. We present algorithms for managing complexity in volumetric rendering algorithms. We present methods for modeling complexity in a volumetric scene. And finally, we present hardware architectures to support these modeling and rendering algorithms. First, we describe our efforts to increase the realism of volume rendering on general-purpose graphics hardware. We propose new algorithms for volumetric global illumination using bi-directional importance methods and novel depth-first, progressive illumination transport. We propose a hardware design capable of simulating general optical transport through volumes enabling direct perspective volume rendering with single scattering, volumetric global illumination, and both parallel and cone-beam tomographic reconstruction. Finally, we give other methods to increase the realism in a volumetric scene through improved modeling. The first improvement is a fast algorithm for accurate, multi-valued voxelization of geometric surfaces. This algorithm is suitable for hardware implementation, and such a design is given. The second improvement is a dynamic scenegraph structure useful for specifying dynamic volumetric scenes of multiple volumes that incorporate the methods described in this work with accurate.