Designing an Inclusive Skin Tone Color Space

This project introduces a custom color space called TUV designed to make digital skin tone selection more intuitive and inclusive. By applying Principal Component Analysis to labeled data, the author created a system that maps skin tones to a spherical space with adjustable diversity. The result is a practical set of equations and a picker UI that balances realism with a broad range of human representation.

Key Points

• Digital skin tone representation often fails by being either too restrictive (like emojis) or too unstructured (like raw RGB).
• Principal Component Analysis (PCA) can transform complex RGB data into a more intuitive coordinate system based on actual human variation.
• The TUV color space provides three distinct axes for skin tone: lightness (T), redness/flush (U), and temperature (V).
• A single tunable parameter, R-squared, allows developers to easily adjust the range of diversity and realism for procedural generation.
• Technical color tools must be developed with an awareness of social contexts like colorism and the historical lack of representation for darker skin.

Sentiment

Predominantly positive and constructive, characterized by technical curiosity and a shared interest in improving digital representation through better math and tooling.

In Agreement

• The TUV space is a significant improvement over the limited color swatches typically found in character creators.
• The use of PCA and function fitting is a clever, effective way to simplify a complex 3D color problem into intuitive controls.
• The project is highly practical for game development, VR avatar software, and procedural generation.
• The author's transparency about manual labeling bias and the inclusion of a 'Humanities Intermission' is thoughtful and commendable.

Opposed

• Manual labeling by a single individual inherently encodes personal bias into the resulting color space.
• At higher R-squared values, the algorithm generates unrealistic skin colors like green, blue, and purple.
• The model may underrepresent the specific genetic expressions of skin tones found in Indigenous Australian or certain East African populations at lower diversity settings.
• A minority of users find the intense focus on categorizing and modeling skin tones to be socially uncomfortable or unnecessary for art.