Balancing Fidelity and Performance in Iridal Light Transport Simulations Aimed at Interactive Applications

Abstract

Specific light transport models based on first-principles approaches have been proposed for complex organic materials such as human skin and blood. The driving force behind these efforts has been the high-fidelity reproduction of material appearance attributes without one having to rely on the manipulation of ad hoc parameters. These models, however, are usually considered excessively time consuming for rendering applications requiring interactive rates. In this thesis, we address this open problem with respect to one of the most challenging of these organic materials, namely the human iris. More specifically, we present a framework that consists in the careful configuration of algorithms employed by a biophysically-based iridal light transport model on the CUDA (Compute Unified Device Architecture) parallel computing platform. We then investigate the sensitivity of iridal appearance attributes to key model running parameters, namely spectral resolution and number of sample rays, in order to obtain a practical balance between appearance fidelity and performance on this platform. The results of our investigation indicate that predictive light transport simulations can be effectively employed in the generation of iridal images that are not only believable, but also controlled by biophysically meaningful parameters. Although our investigation is centered at the human iris, we believe that it can be viewed as a proof of concept, and the proposed configuration strategies and parameter space explorations can be employed to obtain similar results for other organic materials.