Is it possible that low pass filtering can be applied to Ray tracing ?

My guess is that since after the algorithm runs we have an image then low pass filtering helps in order to prevent aliasing from happening. Yet, I am not sure about this. Can someone elaborate on that?


Low-pass filtering is a classic tool from signal theory that will effectively remove noise, as you suggested, but will also cancel out desired high-frequency information in the image such as sharp edges. The image will look blurry.

Post-filtering Monte Carlo rendering results is an open field of research and many advances have been made over the years, and overview of which can be found in the Denoising Your Monte Carlo Renders: Recent Advances in Image-Space Adaptive Sampling and Reconstruction SIGGRAPH Course here.


David Kuri's answer yields a modern approach, but a simple-to-implement solution is to explicitly supersample with jittering. Classic paper: http://www.cs.cmu.edu/afs/cs/academic/class/15869-f11/www/readings/cook86_sampling.pdf.

  • $\begingroup$ Note that there is a fundamental difference: supersampling (with jittering or any other technique) requires additional samples, while a post-filter (like the suggested low-pass filter) works on the rendered image. Nevertheless, this is valuable additional information and both techniques go hand in hand. $\endgroup$ – David Kuri Jan 25 '16 at 12:47
  • $\begingroup$ I answered initially without seeing yours (antialias with some kind of stochastic supersampling), but your answer was better. I still thought the older paper was still worth referencing, so I changed my answer instead of deleting it. I hope that's reasonable etiquette? $\endgroup$ – Daniel M Gessel Jan 25 '16 at 17:34
  • $\begingroup$ Totally, your answer is a fine addition :) I just wanted to make sure nobody gets confused. $\endgroup$ – David Kuri Jan 26 '16 at 8:43
  • $\begingroup$ Understood. It really sinks in that aliasing is a result of the initial samples being at regular intervals: if every sample hits the pickets of a fence, you're stuck with a white wall. Applying a low pass filter to a regularly sampled image just isn't universally effective (the high frequency picket fence aliases down to a low frequency artifact - a white wall). $\endgroup$ – Daniel M Gessel Jan 26 '16 at 14:00
  • $\begingroup$ The post filtering approach also seems like a really interesting modern CPU (with integrated GPU) load balancing question. Ray trace using CPU cores, which walk such data structures efficiently, post process the images on the GPU. Fun project! $\endgroup$ – Daniel M Gessel Jan 26 '16 at 14:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.