Google Starts Work on New Skin Tone Classification System

Google has revealed that it is currently trying to develop a replacement for the Fitzpatrick Skin Type (FST) scale. The FST first became popular in the 1970s, when it was used to inform research into skin cancer and other medical conditions. It has since been used to gauge the accuracy of biometric systems (including facial recognition and heart rate detection systems) that take the subject’s skin tone into account.

The problem, according to critics, is that the FST scale does not adequately reflect the full spectrum of human skin. The scale only includes six colors, four of which are assigned to different shades of white. The other two colors are black and brown, two extremely broad categories that cover a much wider array of different shades and tones.

As a result, many researchers now believe that the limited FST is contributing to racial bias, since it does not recognize the true diversity of people of color. Google is simply acknowledging the problem, and in the process becomes the first of the major tech companies to confirm that it is working on an alternative classification system. The company believes that a more inclusive system will mitigate racial bias and improve the accuracy (and the performance) of its various biometric solutions for people all over the world.

“We are working on alternative, more inclusive, measures that could be useful in the development of our products, and will collaborate with scientific and medical experts, as well as groups working with communities of color,” Google said in a statement.

In that regard, it is worth noting that Google is speaking to a larger trend within the tech industry. The US Department of Homeland Security suggested that FST should be abandoned during a standards conference in October, while Facebook researchers warned about the limitations of the scale in an April study that examined the efficacy of deep fake detection systems. Facebook still used FST tags to label a video dataset of 3,011 individuals, but has indicated that it would support more representative measures. The dataset itself was put together specifically to reduce the amount of bias in face and voice recognition algorithms.

As it stands, the FST’s six-color scale is far more restrictive than the scales used in other industries. Crayola, for instance, offers 24 skin tone crayons, while the Barbie Fashionistas line includes dolls with nine tones.

AnyVision also praised Google’s decision to develop its own solution. The company has repeatedly emphasized the importance of unbiased facial recognition systems, and recently asked the NIST to draft guidelines for the ethical use of the technology.

Source: Reuters

–

June 21, 2021 – by Eric Weiss