Fast Gaussian blur + Image quantization = Fast Gaussian quantization.
This works followed closely the work I have done on image voronoization. The Codegolf thread that initiated the aforementioned project links another related Codegolf thread "Paint by Numbers", which I again wanted to try.
The answer provided by DavidC is a Mathematica code that perform a Gaussian blur step then an image quantization. Pretty simple and giving nice results ! I decided to port the exact same routine to C++, using my implementation of the fast Gaussian blur approximation by IvanK (see details here and my code here) and a classic K-Means pass for image quantization.
Here are the the results with various blur radiis (R) and number of colors after quantization (K):
| R\K | 4 | 16 | 64 | 256 |
|---|---|---|---|---|
| 0 |  |
 |
 |
 |
| 1 |  |
 |
 |
 |
| 2 |  |
 |
 |
 |
| 4 |  |
 |
 |
 |
| 8 |  |
 |
 |
 |
| 16 |  |
 |
 |
 |