Progressive Light Path Building

Abstract

The paper proposes a global illumination method that builds up the light paths progres-sively taking into account all relevant previous information. The basis of the method is a self-correcting stochastic iteration scheme, which works with a population of photon hits. In each iteration step a ray is generated randomly either from a light source or by reflect-ing an earlier hit, then the ray is traced to obtain a new hit. In order to limit the size of the hit population, hits are decimated randomly after certain iteration steps. Comparing the new approach to random walk techniques, this method can reuse the illumination and visibility information gathered with previous rays. By defining the decimation strategy properly, the view-importance can be built into the algorithm.