Item

Abstract

A novel view-independent technique for progressive global illumination computations has been developed that uses prediction of visible differences to improve both efficiency and effectiveness of physically-sound lighting solutions. The technique is a mixture of stochastic (density estimation) and deterministic (adaptive mesh refinement) algorithms that are used in a sequence optimized to reduce the differences between the intermediate and final images as perceived by the human observer in the course of lighting computations. The quantitative measurements of visibility were obtained using the model of human vision captured in the Visible Differences Predictor (VDP) developed by Daly \cite{Daly93}. The VDP responses were used to support 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 the image quality. Also, the VDP was used to determine the optimal sequential order of component-algorithm execution, and to choose the points at which switch-over between algorithms should take place. As the VDP is computationally expensive, it was applied exclusively at the stage of design and tuning of the composite technique, and so perceptual considerations are embedded into the resulting solution, though no VDP calculations are performed during the lighting simulation. The proposed global illumination technique is also novel, providing at unprecedented speeds intermediate image solutions of high quality even for complex scenes. One advantage of the technique is that local estimates of global illumination are readily available at early stages of computations. This makes possible the development of more robust adaptive mesh subdivision, which is guided by local contrast information. Also, based on stochastically-derived estimates of the local illumination error, an efficient object space filtering is applied to substantially reduce the visible noise inherent in stochastic solutions.