Irregular Shadow Maps

Shadows provide important visual cues, increasing the viewer’s ability to accurately interpret intra-object and inter-object spatial relationships in images of computer-generated scenes [8]. Moreover, shadows dramatically increase the ”realism” of synthetic scenes. We propose a new approach to computing shadows based on the image-space technique known as ”shadow mapping”. Our approach builds one shadow map per light source, composed of one sample (or a small constant number of samples) per pixel in the eye view image plane. Samples are irregularly spaced throughout the map. Each sample corresponds exactly to the position of the respective eye view pixel in world coordinates at the Z value given by the depth buffer, from the view of the light source. Our algorithm combines the image quality and sampling characteristics of ray traced shadows, with the performance advantages of depth-buffer based hardware pipelines. We compare the image quality, memory utilization, and sampling characteristics of this approach with ray traced shadows [6], and two existing shadow mapping techniques [5, 9]. CR