标题: Frequency based Kernel Estimation for Progressive Photon Mapping [打印本页] 作者: 彬彬 时间: 2012-1-3 10:44 标题: Frequency based Kernel Estimation for Progressive Photon Mapping We present an extension to Hachisuka et al.’s Progressive Photon
Mapping (or PPM) algorithm [Hachisuka et al. 2008] in which
we estimate the radius of the density estimation kernels using frequency
analysis of light transport [Durand et al. 2005]. We predict
the local radiance frequency at the surface of objects, and use it to
optimize the size of the density estimation kernels, in order to accelerate
convergence. The key is to add frequency information to
a small proportion of photons: frequency photons. In addition to
contributing to the density estimation, they will provide frequency
information.
1 Algorithm Overview
Our algorithm works as follow: First, we ray trace hitpoints in the
scene from the camera. Second, we iteratively trace a proportion
of frequency photons and ”classical” photons in the scene. These
frequency photons are used to update the frequency estimate at hitpoints.
Classical progressive photon mapping progressively reduces the
size of the density estimation kernel to cancel the bias of density
estimation, at the expense of increasing variance. However, if the
radiance is smooth enough, this size can be kept constant, and additional
photons will only decrease variance of the estimate. The
role of frequency photons is precisely to conservatively predict this
phenomenon.
Consequently, we update the hitpoints’ collecting size using the following
heuristic: if the kernel size of classical PPM is greater than
our prediction we decrease the radius just as the standard progressive
photon mapping method does. Otherwise, we don’t decrease
the kernel size. We make this process iterative by always using a
fixed proportion of frequency photons to update the desired size of