Creating a Color Palette from an Image

2 days ago

First iteration used median-cut quantization in RGB with fixed ROYGBIV regions, but it became complex and was replaced by OKLCH and K-means for better color perception.
Second iteration clusters pixels in OKLCH using K-means++ (K=10 overshoot), merges similar colors, and picks highest-chroma pixels as swatches, aiming for five-color palettes.
Third iteration adjusts K to 14 for better accent detection and weights hue/chroma twice as much as lightness during merges to prioritize human color perception.
Fourth iteration introduces structural changes: dropping low-weight/low-chroma clusters, allocating slots by achromatic/chromatic mass, and refining swatch selection based on centroid chroma.
Final algorithm includes phantom guard, slot allocation by mass, and representative pixel selection to produce more human-like palettes without hard-coding edge cases.

Hasty Briefsbeta