The Color of Noise (2014) | Hacker News Reader

The Color of Noise (2014)

caseymuratori.com

80 pointsekzhang12 comments

"In fact, people have actually made spatial pattern generators that allow you to input the frequency profile that you want, and get the corresponding point pattern out. It’s really quite neat, and I highly recommend reading this paper so you can see some other possible noise parameters, like anisotropy."

I had to hunt this 'custom colour' noise paper out since that link was dead. It is quite neat. Here it is on ACM in case anyone else is interested:

Point Sampling with General Noise Spectrum (2012) https://dl.acm.org/doi/10.1145/2185520.2185572

Non-paywalled version on one of the authors' site:

https://www.liyiwei.org/papers/noise-sig12/

The same algorithm is described more clearly in Bridson's paper from 2007:

https://www.cs.ubc.ca/~rbridson/docs/bridson-siggraph07-pois...

Something sounds odd. If points are very close, they have high spacial frequency. If I reject points that are too close, I'm excluding high-frequency points and keeping only low-frequency ones. A distribution that favor lower frequencies and filters out higher ones is red, not blue. Am I getting it wrong?

I think the color terms and frequency may make sense in reference to a comparison against uniformly-spaced points such as a rectangular grid. If you perturb uniformly-spaced points with high-frequency/short-wavelength adjustments, you won't get the large-scale clumping that results from also including low-frequency/long-wavelength distortions.

I had this question and needed this answer - thank you

I was looking through the source code for Bevy's Screen-Space-Ambient-Occlusion (SSAO) implementation and noticed they are using blue noise as well:

https://github.com/bevyengine/bevy/blob/56d559102858d4ce8a5b...

That link takes you to this shadertoy:

https://www.shadertoy.com/view/3tB3z3

It is a common pattern in rendering. You achieve your desired results, but it is nowhere near realtime. You try to lower the sampling resolution which helps a lot with speed, but you end up with ugly visual artifacts. Introducing the blue noise in the right place helps with attenuating these sampling artifacts. Only usually you would sample the premade "blue noise" texture rather than using the shader to compute it.

This article does a good job of explaining how it helps in raymarching:

https://blog.maximeheckel.com/posts/real-time-cloudscapes-wi...

And for an IMO much better explanation how it can be used more or less everywhere, this presentation by Playdead games on how it's used in INSIDE: https://www.youtube.com/watch?v=RdN06E6Xn9E

IMO the linked article by Casey Muratori is actually kind of weak: a lot of waffling, doesn't actually show the results of using this point set in the original problem, just using the sampling pattern isn't enough to eliminate aliasing, and there are more important properties for point sequences to have (discrepancy), etc. It's somewhat unpopular these days to link dense texts instead of "softer" articles, but IMO one should just go straight to PBRT for this topic, there's an entire chapter with excellent presentation: https://pbr-book.org/4ed/Sampling_and_Reconstruction

Final note: the Don Mitchell mentioned in this article is one of the authors on the go-to reconstruction filter used in a lot of software, Mitchell-Netravali (the latter recently passed away): https://en.wikipedia.org/wiki/Mitchell%E2%80%93Netravali_fil...

I would be very curious to hear if anyone knows how the color names are assigned?.. Is there any link to frequencies/spectra of light or is it kinda arbitrary?

the former: https://news.ycombinator.com/item?id=42291527 https://news.ycombinator.com/item?id=42292668

And here I was thinking it was because noise space is non-riemannian

Reply on news.ycombinator.com