Using the visual differences predictor to improve performance of progressive global illumination computation

Abstract

A novel view-independent technique for progressive global illumination computing that uses prediction of visible differences to improve both efficiency and effectiveness of physically-sound lighting solutions has been developed. The technique is a mixture of stochastic (density estimation) and deterministic (adaptive mesh refinement) algorithms used in a sequence and optimized to reduce the differences between the intermediate and final images as perceived by the human observer in the course of lighting computation. The quantitive measurements of visibility were obtained using the model of human vision captured in the visible differences predictor (VDP) developed by Daly [1993]. The VDP responses were used to support the selection of the best component algorithms from a pool of global illumination solutions, and to enhance the selected algorithms for even better progressive refinement of image quality. The VDP was also used to determine the optimal sequential order of component-algorithm execution, and to choose the points at which switchover between algorithms should take place. As the VDP is computationally expensive, it was applied exclusively at the design and tuning stage of the composite technique, and so perceptual considerations are embedded into the resulting solution, though no VDP calculations were performed during lighting simulation. The proposed illumination technique is also novel, providing intermediate image solutions of high quality at unprecedented speeds, even for complex scenes. One advantage of the technique is that local estimates of global illumination are readily available at the early stages of computing, making possible the development of a more robust adaptive mesh subdivision, which is guided by local contrast information. Efficient object space filtering, also based on stochastically-derived estimates of the local illumination error, is applied to substantially reduce the visible noise inherent in stochastic solutions.