--verbose: Bring the (Perlin) Noise

If you just want the source code: The JavaScript Perlin noise generator code is here.

I mentioned in a previous post that I was working on a Perlin noise generator for Art Evolver.

Perlin noise is a function of (x, y) that produces a random-ish pattern. It's not completely random because it has smooth hills and valleys, but the distribution of those hills and valleys is random.

A side note about this algorithm: Usually if you're a computer scientist and you come up with a clever algorithm to solve a particular problem, you get an award from a university, or a CS-centric professional organization like the ACM or IEEE. Perlin got an Academy Award for this noise function. As in, the Oscar kind of Academy Award. For an algorithm, something not usually consider artsy. I found that interesting.

Anyways, rather than dive into Perlin's impenetrable description of how the algorithm works, I set out to find an existing JavaScript implementation. That let me to this message board, and specifically this example.

Unfortunately that implementation has some pretty serious directional artifacts:

Note the horizontal and vertical stripes. There's almost an upside down cross in the lower right. SAAAATAAAAN!

Rather than try to fix that source code (which the author apparently closureized (making it very difficult to understand)) I kept searching.

From the main Wikipedia entry on Perlin noise, I ran across a variant called Simplex noise. This is an improvement on the original algorithm, also written by Ken Perlin, in 2001. That Wikipedia page linked to a paper by Stefan Gustavson(pdf) that explains both classical and Simplex Perlin noise in a much easier to grok way than anything I've read by Perlin himself. I highly recommend Gustavson's paper if you found Perlin difficult.

I took the Java source code in Gustavson's paper and ported it to JavaScript, and the results are here on github.

I ran some performance comparisons between the classical and Simplex algorithms, and for 2-D I only saw a ~10% improvement with Simplex. Granted, the latter is supposed to be faster in higher dimensions (classical is O(N^2) vs. simplex O(N) where N is the number of dimensions) so it doesn't matter much for my purposes.