‘Dazzle’ makeup won’t trick facial recognition. Here’s what experts say will

As demonstrators protest against racism and police brutality, some have suggested that extravagant makeup can block facial recognition technology they worry have been deployed by authorities.

But the creator of this “CV Dazzle” makeup style said the patterns, which were designed to fool an older method of facial detection, won’t trick more sophisticated algorithms — though he and other experts said protesters can take steps to evade detection.

Tear gas and rubber bullets aren’t the only risks for those who attended protests over police violence and the death of George Floyd. Protesters also stand the risk of having their identity revealed by the myriad of facial recognition tools available to police departments and government agencies in the U.S.

The images of thousands of faces of protestors have been captured on social media or by news organizations since protests began in late May. OneZero reported that local law enforcement, as well as the FBI, have asked the public for footage of the protests in order to identify criminal activity.

In the face of having their identity disclosed, protesters are seeking a way to guard their anonymity. Some have turned to “CV Dazzle” makeup.

“CV Dazzle” is a Cubist-inspired style of makeup using contrasting shades of angular shapes and unconventional hairstyles that was designed to thwart surveillance systems. “CV” refers to computer vision while “dazzle” references a style of camouflage used in World War I to disguise ships.

Instructions on how to apply this anti-surveillance makeup surfaced on social media last week. One tweet included a screenshot of makeup tips that advise protesters to “partially obscure” their nose bridge area and use colors that contrast with their natural skin tone.

CV Dazzle has inspired makeup artists to experiment with the technique.

Martayla Poellinitz, a Pittsburgh-based makeup artist who posts tutorials under the handle MartyMoment on Instagram, was celebrating her birthday when she heard about Floyd’s killing. She told Digital Trends that she spent the whole day grieving and resolved to do something with her platform. She learned about the CV Dazzle technique on another friend’s Instagram Story, and decided to do more research.

Poellinitz said her first attempt at CV Dazzle didn’t fool her iPhone’s FaceID. She tried applying more makeup on the second try, but that effort failed too. Finally, on the fifth try, after adding press-on gems, she reached a point where she was able to fool her phone’s scanning system.

Poellinitz posted a video on Instagram documenting her experiments to use CV Dazzle to fool an iPhone’s FaceID system, and noted in the caption that it was an experience in trial-and-error.

“Technology gets harder to fool every day,” she noted.

View this post on Instagram

Computer Vision Dazzle or CV Dazzle is a method used to obstruct facial recognition algorithms from identifying your face. In this day and age, when a simple snapshot of your face can expose so much information about you and others you associate with, it’s important to stay as safe as possible. ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ From trial and error, I’ve come up with these conclusions: ⠀⠀⠀⠀⠀⠀ (1) technology gets harder to fool everyday ⠀⠀⠀⠀⠀⠀ (2) The systems need only see a fraction of ONE of these facial key points: eyes, nose and the mouth to identify where other parts of the face may be ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ (3) Jewels seem to work the best for obstruction because of how light reflects off of them ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ Key points: ⠀⠀⠀⠀⠀⠀ (1) systems recognized me with makeup AND wearing an eyepatch ⠀ (2) Masks that cover nose/mouth do not work because your eyes are visible ⠀⠀⠀⠀⠀⠀ (3) If you’re black or POV, you will have an easier time tricking technology, because algorithms have racial bias ⠀⠀⠀⠀⠀⠀ (4) do not use oil based makeup, it’ll bond with teargas. ‪Water based and non-toxic acrylic are the way to go ‬ ⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀ Inspo for final look: @laramiemakeup for @dazedbeauty #cvdazzle #antisurveillancemakeup #antifaces

A post shared by Martayla ???? (@martymoment) on

Adam Harvey, the Berlin-based artist who coined the term “CV Dazzle” as part of his master’s thesis at New York University, warned that the designs for anti-surveillance makeup, which are now a decade old, are no match for today’s advanced “deep convolutional neural networks.”

Digital Trends initially reached out to Harvey for comment, who referred a reporter to a tweet on the topic. Harvey later contacted Digital Trends to clarify that the algorithm the initial dazzle makeup was based on “has since been deprecated.”

Harvey also noted that the “CV Dazzle” he referred to was not the style of makeup, but the process itself. Harvey said his method of developing the makeup, which involves reverse-engineering algorithms and developing looks to fool them, could still be applied to modern systems.

Harvey’s initial makeup designs may still fool your iPhone’s FaceID, but they are no match for more advanced programs used in investigations. The trouble with disguising yourself to fool facial recognition systems is that the term “facial recognition” describes a broad range of tools and programs that vary in success rate.

Unlike automated facial recognition used by FaceID, law enforcement agencies will crop a photo of the person they wish to identify, and run it against a database of photos, such as mugshots, in order to surface a match, experts said.

“Deep face recognition systems have been shown to be highly robust to facial occlusion or masking,” Dr. Christian Rathgeb, a senior researcher with the Faculty of Computer Science in Hochschule Darmstadt, Germany, told Digital Trends.

If you have a clean criminal record and your local police department only runs your photo against a database of mugshots, you could get off scot-free. But if your city uses Clearview A.I., a facial recognition vendor that scrapes social media photos, it could trigger a match.

Depending on the protestor’s visibility, camera footage or smartphone photos could prove insufficient, said Anil Jain, a computer science professor at Michigan State University.

“A protestor may not be visible in all of the CCTV cameras,” Jain told Digital Trends. “[Authorities] could ask the public to share images of the protestors which may have been captured on the smartphones of the bystanders. But, given the crowd density, that may be problematic.”

While masks may not be foolproof, they may be the best option protesters have right now, experts said.

“You can just wear a mask now,” Harvey said in his tweet. “It’s much easier.”

Jain agreed that wearing a face mask or a bandana to cover parts of your face is sufficient to throw off most face recognition systems.

Rachel White, a senior software engineer based in Brooklyn who has advised protestors, suggests covering up birthmarks or any identifying features.

“Normal black masks and sunglasses, as we have them now, offer the best widespread protection, as we don’t know for sure what the police are using,” White told Digital Trends.

But companies around the world are already working on technology that can identify individuals with masks or head coverings.

Hanwang, a Chinese facial recognition vendor, has already debuted cameras in China that can identify masked individuals, according to ArsTechnica.

In Hong Kong, widespread uncertainty over how and where the government is using facial recognition has forced protestors to take many creative measures, including wearing full-face masks. Some have cut down smart lamp posts while protesting, The Atlantic reported.

No one is certain what kind of technology law enforcement agencies are using at any time, or may use in the future. But at present, a mask appears the best option for protestors who wish to evade detection.

Correction: This article initially misinterpreted Harvey’s tweet about CV Dazzle. Harvey later clarified to Digital Trends that the algorithms the initial style was based around had “deprecated”

Want more news, reviews, guides, and features from Digital Trends? Follow us on Apple News, Google News, and Flipboard.

Editors' Recommendations