Adaptive light paths generation through full lens model

Abstract

Photo-realistic image synthesis using full lens model provides us with realistic lens effects, but it suffers from low rendering efficiency when many light paths are obstructed by lens stops and lens barrel. This paper proposes a novel method to generate light paths, along which rays can propagate through the lens system. As a first step, a light passage function is defined as the objective function for sampling light paths. The sampling is implemented in a hypercube space, by means of both adaptive Markov chain sampling and interacting Markov chain Monte Carlo. Then light paths are constructed based on these samples. This approach can be easily incorporated in existing rendering methods to trace rays through a full lens model. Experimental results show that this approach can effectively increase the number of valid rays which can go through the lens system, therefore improving the rendering efficiency.