The most common shading algorithms in direct volume rendering are based on lighting models known from surface-based rendering. Depending on the material of the volume dataset, the resulting images may not look realistic but have a metallic look instead which is caused by neglection of most light-matter interactions in basic lighting models. More sophisticated lighting models do not only result in more realistic renderings but also can improve the image quality and visualization itself by revealing fine details and providing depth cues that help to understand spatial relationships.

This thesis presents a lighting model for GPU-based ray-casting that considers volumetric shadows as well as an approximation to scattering. Further, the existing framework was extended to support high dynamic range rendering. The resulting images look more realistic and have a smooth appearance which makes
them visually appealing.

GPU-Accelerated Deep Shadow Maps for Direct Volume Rendering (running)

Renderings of a CT scan of an engine (256x256x110). Shadows provide an enhanced depth perception and a better understanding of spatial relationships.

The problem with shaded DVR (left) is, that in homogeneous regions the gradient is undefined which leads to noisy images.  Shadows (right) provide an image that has a shaded appearance without the need of any gradient computations.

Example rendering of a CT scan of a human hand (256x256x128). The tone mapping function keeps highlights as well as shadows. Light regions are compressed more than dark regions.

Rendering of a CT scan of a human torso (512x512x1112) with DVR and shadowing with GPU-accelerated deep shadow maps on a 512 MB graphics card.