Spatio-Temporal Photon Density Estimation
Using Bilateral Filtering 

Markus Weber, Marco Milch, Karol Myszkowski,
Kirill Dmitriev, Przemyslaw Rokita, and Hans-Peter Seidel

MPI Informatik



For HDR images and videos captured in the real-world information corresponding to luminance levels for each pixel is only available (camera calibration may be required to recover absolute luminance levels). Given such limited information the most general image and video processing algorithms rely only on pixel values taking into account their local neighborhood in space and time (camera and object motion compensation may be required in the latter case). Recovering lightness in such an input material as might be required for luminance compression (tone mapping) or object appearance prediction is an under-constrained problem whose solution may lead to some errors. For computer generated images information about lighting (direct, indirect, caustic paths, and so on), reflectance is explicitly available but its quality might be poor. For example stochastic noise in lighting (which is a HDR signal) is common in the global illumination computation. Also, BRDFs describing surface reflectance can be poorly sampled and some high frequency information may be poorly reconstructed. Then different strategies applied to filtering each of those information layers might be desirable.


We present a method for efficient filtering of noise in indirect lighting distribution inherent for Monte Carlo methods used for the global illumination computation. For this purpose we use spatio-temporal bilateral filtering of photon density over the meshed scene surfaces. Bilateral fileting enables control over the level-of-details in reconstructed lighting. All changes of lighting below this level are treated as stochastic noise and are suppressed. Bilateral filtering proves to be efficient in preserving sharp features in lighting which is in particular important for high-quality caustic reconstruction. Also, flickering between subsequent animation frames is substantially reduced due to extending bilateral filtering into temporal domain.

This work   (PDF,  730 kB) has been presented at Computer Graphics International 2004 (BibTeX Entry).



Möbel Martin Sala

  DivX;-> MPEG 2 Quicktime
320 x 256 5.7 MB    
640 x 512 7.3 MB    
  DivX;-> MPEG 2 Quicktime
320 x 256 0.5 MB    
640 x 512 1.3 MB    
  DivX;-> MPEG 2 Quicktime
320 x 256 0.9 MB    
640 x 512 1.8 MB