Atima Lui was in primary school when she first learned that “nude” is not universal. Now 30, she still recalls playing with a white friend’s makeup and struggling to find colours that complemented her rich skin tone. “I would try to put [her makeup] on and it would just make me look like a clown,” says Lui, who is of Sudanese and African-American descent. “I think back to growing up and how my mother barely wore makeup. Now I know it’s because makeup just wasn’t made for her.”
The cosmetics landscape has long been unfriendly terrain for anyone on the wrong side of beige. Before Rihanna introduced her groundbreaking Fenty Beauty line with 40 shades of foundation in 2017, pushing competing brands to diversify their palettes or face public backlash, people with darker skin had few accessible options that matched and enhanced their complexion.
But what Rihanna has done to address the issue of foundation shade selection, Lui is now hoping to do for colour matching – finding the perfect shade of makeup is usually left to guesswork or performed by associates on the beauty department floor. With her computer vision tool, Nudemeter, users simply upload a selfie and complete a short quiz, and an algorithm suggests the product that best matches their skin tone.
Lui first conceived of the idea in 2016, during her final year at Harvard Business School, as a tool to empower dark-skinned shoppers to make purchases that helped boost their confidence. “I just went back to being a Black woman growing up in Topeka, Kansas, and just not feeling beautiful, not feeling like the standard of intelligence, not feeling good enough,” she explains. “Beauty is undervalued as a source of power in the world.”
But developing Nudemeter was no easy undertaking. The world of facial recognition technology is as guilty of light-skin bias as the beauty industry. A 2018 MIT study, led by Algorithmic Justice League founder Joy Buolamwini, found that commercial artificial intelligence systems had error rates as high as 35 per cent when identifying the features of darker-skinned women, compared to less than one per cent for lighter-skinned men – a discrepancy attributed to datasets “overwhelmingly composed of lighter-skinned subjects.”
To avoid this problem, Lui had to train her algorithm with images that more accurately represented the skin-colour spectrum, from the palest whites to the darkest browns. To this end, she issued call outs encouraging volunteers of all skin tones to submit photos of themselves to aid in her mission to “change the standard of beauty to match the full range of diversity in human skin.” Once she had a dataset in place, she reached out to Michael Brown and Mahmoud Afifi at York University in Toronto, who specialise in color analysis and digital image processing, to make sure the algorithm could deduce the user’s true skin tone, regardless of their device or the conditions in which their photo was taken.
“[Our phones] are really intended to create beautiful images, not images with color accurate measurements… which makes the challenge that I’m working on that much more difficult,” Lui says. “It’s all about using AI to predict the color of the real scene that is depicted in the image, and not actually measuring the color from the pixels.”
The potential for this technology hasn’t gone unnoticed. In 2018, beauty behemoth Coty, whose brands include Rimmel, Max Factor and Kylie Cosmetics, awarded Nudemeter the grand prize for their Digital Accelerator Start-Up Program, and helped Lui refine and stress test her algorithm. Last year, Spktrm Beauty, an independent brand targeting shoppers with darker skin, became the first to utilise Nudemeter on its website, and in May, hosiery company Nude Barre introduced the app to help shoppers pick out the right tights for them.
Looking to the future, Lui hopes to see further growth on the colour-matching side, but also sees potential for her technology beyond that. “I think there’s power in using it for opportunities like virtual makeup try-on, or virtual glasses try-on, or even improving Instagram filters,” she says. “It’s about feeling seen, feeling beautiful and having fun.”
She also envisions someday sharing her dataset (which is proprietary) with other companies attempting to create more inclusive AI technologies and combat existing biases. But in the meantime, her priority is refining and challenging the Nudemeter algorithm to be as inclusive as possible.
“I’m really proud of how well my technology can measure the skin tones, undertones and differences of dark-skinned women… But how is the technology reading the faces of people with vitiligo? What about people who are bald and have hair loss? What about people who are over 70 years old and might have a lot of wrinkles?” she says. “This work of creating inclusive and representative technology is never done.”