A parallel image generation algorithm based on Photon map partitioning

Photon mapping attracts much attention as an excellent image generation technique that can simulate various lighting effects such as indirect illumination and caustics obtained only by the global illumination model. Although photon mapping can generate high-quality images, it requires more expensive calculations and a large memory capacity. In this paper, we present a new parallel photon mapping algorithm to solve the problems regarding the computing time and memory requirement. The proposed algorithm can effectively parallelize photon map construction and photon search by distributing partial photon maps among processing elements of a parallel computer. As a photon map is partitioned, only a part of the photon map is assigned to each processing element. Therefore, each processing element does not require a large memory space even if the entire photon map is quite huge. We implement the proposed algorithm using MPI and evaluate it through experiments on a parallel computer. The experimental results indicate that our algorithm can significantly reduce the rendering time of photon mapping as the number of processing elements increases, and can also save the memory space.