Perception-Based Techniques for Improving Computer-Generated Imagery

摘要

The current program is focussed upon the development and evaluation of techniques that reduced the computational resources required to achieve specific levels of graphical fidelity in computer generated imagery (CGI) systems. These techniques are based upon perceptual principles and allow a given level of apparent realism to be achieved at a reduced computational cost. The basic idea underlying all of these techniques is that since the human perceptual system neither uses nor requires all of the optical information available in a scene when forming spatial perceptions, efficiency gains can be realized in CGI systems by requiring that they render only information that is of perceptual utility. There were three primary areas of research. The first area examines the feasibility of exploiting aspects of visual fusion processes in order to increase the apparent resolution of images at a lower computational load than is required by current techniques. The second area examines the use of perception-based techniques to automate the modulation of level of detail in time-critical rendering. The third area focuses on developing techniques to streamline efficiency of animation systems by exploiting the fact that the perceptual system uses only a subset of the available optical information when deriving three-dimensional form from motion. Generalizing the principles of visual fusion to audition, a technique was developed for decreasing the computational load required to generate and transmit stereo auditory information.