What is it about?
In fields ranging from medicine to AI research, technology developers often label the skin tone of a person in a photograph using charts or scales. However, the way these labels are determined can be inconsistent or inaccurate, which can lead to incorrect findings or technology that is less accurate or reliable. In this study, we developed a new skin tone scale based on precise color measurements from real people. We tested how well people could use this new scale—both to describe their own skin and to rate skin tone in photographs. Compared to older scales, our new scale provided more accurate and consistent results. This can improve the quality of data used in technology and research that relies on skin tone information.
Featured Image
Photo by Clay Banks on Unsplash
Why is it important?
Creating computer vision technology that works reliably is important. One factor that we know impacts computer vision algorithms that operate on photographs of humans is the skin tone of the person in the picture. But how do we measure this? Our study is the first to directly compare skin tone rating scales to actual measured skin color using color science. While existing scales are widely used, their accuracy had not been validated with standardized color measurements. We developed a new color-based scale and showed that it performs better across a demographically varied set of individuals and conditions. This work provides a more accurate method for describing skin tone, which is increasingly important as skin tone is used in photography, research, and technology development involving humans.
Perspectives
There has been long-standing uncertainty in the field about how best to characterize human skin tone. Some past work has adopted the Fitzpatrick scale—originally developed for dermatology—which unfortunately lacks any defined color values. Other efforts have used color scales with unknown origins or justification. We began measuring skin tone to better understand demographic variation in biometric and identity systems. It was exciting to build on that work to create a repeatable, data-driven method for developing a colorimetric scale grounded in real skin tone measurements. For me, this project reinforced how iterative refinement of measurement tools can enhance the quality of applied research—and ultimately support the development of more accurate and reliable technologies.
Yevgeniy Sirotin
Read the Original
This page is a summary of: Colorimetric skin tone scale for improved accuracy of human skin tone annotations, ACM Journal on Responsible Computing, June 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3730409.
You can read the full text:
Contributors
The following have contributed to this page
