Bias compensation for photon maps

Density estimation techniques such as the photon map method rely on a particle transport simulation to reconstruct indirect illumination, which is proportional to the particle density. In the photon map framework, particles are usually located using nearest-neighbour methods due to their generality. However these methods have an inherent tradeoff between local bias and noise in the reconstructed illumination, which depends on the density estimate bandwidth. This paper presents a bias compensating operator for nearest-neighbour density estimation which adapts the bandwidth according to the estimated bias in the reconstructed illumination.