Okmain: How to pick an OK main colour of an image

5 days ago

The article discusses a method to extract a visually pleasant, representative solid color from an image for the back of a card.
A common approach of resizing the image to 1x1 pixel often results in dull colors, prompting the development of a better solution.
The author created a library called Okmain, which uses k-means clustering in the Oklab color space to find dominant colors.
Okmain limits clusters to four for performance and quality, adjusting if clusters are too similar.
Operating in Oklab avoids issues with sRGB, such as non-linear gamma correction and perceptual color intensity, leading to smoother color mixing.
Visual prominence of colors is determined by cluster size, centrality in the image, and chroma (saturation).
Optimizations include downsampling images, using SIMD for performance, and focusing on auto-vectorization-friendly code.
The author experimented with LLM agents for development but found manual rewriting necessary for clarity and correctness in critical parts.
Despite challenges, the project resulted in a functional library, with the author learning about k-means, SIMD, and mixed Python/Rust libraries.

Hasty Briefsbeta