Soft shadow visualization in synthetic scenes.

다국어 초록 (Multilingual Abstract)

This dissertation presents two two-stage soft shadow visualization methods for area light sources. The first method, visualizing soft shadows for deformable moving area light sources, uses adaptively sampled shadow maps to allow viewers to rapidly modify the shape and the location of area light sources and visualize the changes of soft shadows in a synthetic scene. During the pre-processing stage, this method creates a shadow slab that represents intermediate shadow information for all light samples and a scene map that represents surface samples captured from a pre-defined viewpoint. To reduce storage cost, our importance-based adaptive sampling technique stores the shadow maps that form the shadow slab and the scene map in quad-trees, and only stores the fine details for essential areas. In the visualization stage, we update the attenuation map and screen buffer whenever light sources are modified in the light template by accessing the shadow slab and the scene map to provide rapid visual feedback.
The second method, forward area light map projection, is suitable for arbitrary shape of area light sources. It projects sampled surface points (stored in a layered area light map) onto the viewing screen to rapidly generate high-quality soft shadow images. The preprocessing stage creates the layered area light map. The map is multi-layered since each map cell stores the visibility ratio (light attenuation value) of the area light source with respect to multiple surface points at various depths. In order to rapidly generate the final image in the soft shadow rendering stage, we forward project light attenuation values of surface points that are stored in the layered area light map onto the screen buffer to attenuate the shadowless reference image. This forward area light map projection renders images much faster than ray tracing and still results in soft shadows with comparable quality.
In addition, we also present the "priority point lists for point-based object rendering" method. This method uses an adaptive sampling technique similar to the one introduced in the first method and the dynamic splatting technique introduced in the second method.