Google announces 'Monk Skin Tone Scale', a new scale that allows AI to evaluate skin color on a 10-point scale
In collaboration with Professor Ellis Monk of Harvard University, Google has announced the ' Monk Skin Tone (MST) Scale ', which summarizes skin tones in 10 levels. This MST scale will be incorporated into various Google products by the end of 2022.
A closer look at the research to help AI see more skin tones
Improving skin tone representation across Google
https://blog.google/products/search/monk-skin-tone-scale/
In 2021, Google announced 'Real Tone', a function for beautifully shooting various skin colors with Pixel 6's camera and AI-based image processing. Below is an introduction movie of Real Tone released in April 2021.
Google Pixel 6: Real Tone-YouTube
The MST scale was developed by Google in collaboration with Dr. Monk to further evolve this real tone. Dr. Monk has been studying how skin color and colorism affect people's lives for more than 10 years, and the MST scale, which is the culmination of that research, is a scale that evaluates skin color on a 10-point scale. And that.
Until now, the ' Fitzpatrick Scale ' compiled by dermatologist B. Fitzpatrick in 1975 has been used to process skin tones on a computer. The Fitzpatrick Scale evaluated skin in six tones, classifying skin pigmentation exposed to UV light, rather than race or ethnicity.
Dr. Monk said, 'People feel that they are grouped into racial categories, but our research shows that there is heterogeneity in ethnic and racial categories. Past skin tones. The method of scaling and categorization does not focus on diversity, 'he said, saying that the traditional Fitzpatrick scale cannot achieve algorithm fairness.
This MST scale will be applied to image processing AI for photos taken with smartphone cameras such as Pixel, and will also be used for Google image search. For example, Google says that searching for makeup-related queries on Google images can now further refine the search results by skin color. This is possible by labeling each image content based on the MST scale.
Google also said it will be able to label content not only by skin color, but also by attributes such as hair color and texture. Google also wants to open the MST scale to the public, make it available to other companies for its products, and get feedback to drive interdisciplinary research.
You can see the movie that introduces the MST scale from the following.
Monk Skin Tone Scale | Google's Research Center for Responsible AI and Human-Centered Technology --YouTube
Related Posts:
in Software, Web Service, Video, Posted by log1i_yk