Cornell University Program of Computer Graphics
Adaptive techniques for hardware shadow generation.Randima Fernando.
Master's thesis, Cornell University, 2002.
This thesis presents two adaptive algorithms for shadow generation. Both algorithms utilize commercial graphics hardware to accelerate the rendering process. The first algorithm, called Adaptive Shadow Maps, deals with removing the aliasing artifacts that typically result when using shadow maps for hard shadow generation. It achieves this by varying the shadow map resolution spatially throughout the scene based on the eye position. The second algorithm, called Adaptive Soft Shadows, attempts to generate soft shadows efficiently by varying the number of samples needed over the different regions of the scene. More samples are devoted to soft shadow regions that subtend a large portion of the image plane, and fewer samples are devoted to hard shadow regions.
We show that Adaptive Shadow Maps enable dramatic improvements in shadow quality while maintaining interactive rates and being constrained to a user-specified memory limit. In the case of Adaptive Soft Shadows, we examine the reasons why the approach was not as successful as we had envisioned and discuss possible avenues for improvement. The motivation for each algorithm is presented along with the corresponding theoretical foundation, implementation, results, conclusions, and directions for improvement.
This paper is available as a PDF file Fer02.pdf (3.5M).