By: Hessie Jones
Over the last decade we’ve witnessed the deluge of online photo-sharing. From Pinterest to Instagram to Snapchat the rise of photo-sharing has enabled the soaring activity of selfies, citizen journalism, and events in-the-moment. Brandwatch’s recent stats revealed among 18-24 year-old’s, over 75% use Instagram and Snapchat, and 94% use Youtube:
- On Instagram More than 40 billion photos have been shared so far
- Over 95 million photos are uploaded each day on Instagram
- In Q3 2017, 3.5bn snaps were sent.
The future of image and video is clear based on these previous estimations:
- An estimated 84 percent of communications in 2018 were visual.
- An estimated 79 percent of internet traffic content in 2018 was video.
- Posts that include images produce 650 percent higher engagement
“We just couldn’t do that [upload photos online freely]. We had a different point of view. We understood what governments were doing with photo technologies even back then. The reality of the global influx of image sharing, with cameras everywhere, with the promised excitement of Google Glass – all combined with emerging face recognition technology, we saw the future. That future is happening today and it is moving towards an increasingly dangerous path.”
Enter D-ID, short for De-Identification. Gil Perry, CEO, Sella Blondheim, COO and Eliran Kuta, CTO realized early on they needed to do something about the implications of facial recognition technology. A programmer by trade, Gil enrolled in computer science at the university six years ago. By that point, he and Sella had already done significant research on facial recognition and the options to protect photos and to store photos. They started building a solution so people could upload photos, use them online but the contents would be unrecognizable. At that time, no one – companies nor investors – were aware of face recognition technology.
Facial recognition technology began as early as 1960. Woodrow Wilson Bledsoe, the purported father of facial recognition technology created the first semi-automated face recognition system known as the Rand Tablet. By the early 1970’s, Goldstein and Harmer and Lesk were able to increase facial recognition’s accuracy by identifying 21 markers such as hair colour and lip thickness. But it wasn’t until the early 1990’s when the Defense Advanced Research Projects Agency (DARPA) and the National Institute of Standards and Technology (NIST) rolled out the Face Recognition Technology (FERET) program to encourage innovation for specific use for intelligence, law enforcement and security. Gil explains,
“We understood that cameras were everywhere. On London Bridge, people are being captured over 300 times a day. Almost everyone has a smartphone, taking pictures, uploading them everywhere without really understanding the risks. So when you combine three factors: face recognition increasingly accurate, a market of photos and images everywhere, and the business storage of photos from customers, employees and visitors – this amounts to a perfect storm. The realization is that privacy is gone. We have lost our basic human right to privacy. Anyone can track you, recognize you and consequently, steal your identity.”
Gil had developed a working prototype and approached one of the top experts in face recognition who had developed one of the first algorithms using Deep Learning. Through a series of encounters he amassed enough knowledge to understand how governments and businesses were using current technologies and Gil, Sella and Eliran decided to bring their solution to the world with the mission to restore privacy or at least stop this imminent future from happening.
“You’re not going to be able to walk the streets without anyone knowing who you are, who your children are, where are they going, who are their friends, and where they live. If we don’t do anything, this will be our future.”
Before long, Sella, Gil and Eliran incorporated D-ID. The date was January 2017. They pitched to VCs and enterprise and it fell on deaf ears. People saw them as paranoid. Their efforts to explain their research and their experience did not convince investors there was a market need. They were told facial recognition didn’t work. The team also realized the market was largely unaware of the dangers of facial recognition technology. They continue, undeterred, to pitch their story and their solution.
“This was a perfect time for us. Our backgrounds gave us unique personal pain points that put us on this path to do something about it. We believed in each other and we continued to develop our tech and make it stronger. No one was thinking the way we were and that was good for us. What kept us going? We just knew!”
And then regulation happened. On May 25, 2018 the European General Data Protection and Regulation (GDPR) launched, the first globally recognized set of laws that fundamentally sought to change data practices in Europe. Its tentacles were far-reaching and started influencing policy and privacy changes globally. The team knew that business would eventually feel its wrath. Under GDPR, facial images are considered sensitive personally identifiable information (PII). D-ID realized they would be able to fill the gap between regulation and the reality of wide-spread image sharing. D-ID planned to be the privacy-enhancing technology for photos and videos.
D-ID uses advanced image processing and deep learning techniques to resynthesize any given photo in such a way that the photo looks similar and good enough to the human eye but looks different to facial recognition algorithms. Gil explains it is not the same photo. It looks like you but it’s actually different. Typical noise reduction filters won’t help to reduce the protection. However, by applying solutions such as GANS attacks (adversarial noise injection), the user also can upload the photo to social networks, print it, take a screenshot – and it will remain protected.
“It is robust enough against reverse engineering attempts that may try to overcome the protection by training a new classifier to ignore D-ID changes.”
FROM PRIVACY TO COMPLETE ANONYMIZATION
D-ID has recently developed another product. Instead of protecting the photo, their Anonymization product changes the face so it is visibly different than the original. While the identity is masked, what remains are important attributes including age, gender, ethnicity, emotion etc. Gil explains,
“Face recognition is everywhere. Companies are trying to understand our intentions, our emotions when we shop and when we connect with others. The reality is that we can’t stop profiling from happening. That is where the money is. Our identities are being used without our consent. However, if companies want to make their product better by aggregating data from millions of people, identity should not be a factor in this research. It is not necessary to go down to the individual level.”
In this market of Deep Fakes, used in impersonations on social networks and hoaxes on dating sites, not to mention the development of people who do not exist, Gil is clear D-ID’s intent is to protect people’s privacy. The risk, however, is for anyone to potentially leverage the technology to perpetuate this market.
“We don’t focus on protection against deep fakes. Our goal is to make our tech so robust no one can re-identify the photo. We are focused on protecting privacy. In tandem with regulation we aim to leverage our technology to preserve individual privacy. Internet giants are experiencing horrific privacy issues today because they were late to understand the implications of their decisions. They need to change their strategy towards privacy to gain the trust back from their users”
D-ID is selling to privacy-sensitive organizations who hold large repositories of photos from employees, customers/patients or visitors and need to comply with privacy regulations such as GDPR. This includes schools, healthcare, financial organizations, governments for compliance, KYC or physical badge or license identification.
D-ID is also selling to B2B2C: companies who handle millions of consumer images, including device manufacturers, cloud providers/security providers and security providers. Photos can be immediately protected through their API. In the future, D-ID can also be installed on phones to protect user photos. Gil references Ntech labs and their now defunct application, Findfacethat allowed you to identify an individual by simply using photo search. D-ID seeks to render these types of applications useless. It is, as Gil contends, a question of market trust.
Gil feels D-ID surpasses current competitor solutions. Competitors in the anonymization space use pixelization to distort the appearance of the photo to the extent that final image is unusable. Parrot recognition is used to determine the degree of pixelization to re-identify the original image with some degree of accuracy. He insists these methods were not meant to protect from face recognition.
“Our view is that you should not have to hide. You do not have to wear glasses or hats to deter others from identifying you. You should be able to freely walk in plain sight while protecting your most sensitive information.’
To date, D-ID has raised $10MM, employs 5 PhDs, and the top deep learning and computer vision experts in Israel. D-ID’s advisors validate the technology and vision:
“Faces have become our digital identifiers and as such they must be strongly protected. As more systems adopt facial recognition, the risk to privacy escalates dramatically. That’s why every organization that handles images – corporations, governments, security agencies – needs to act quickly and implement powerful security, deidentification and privacy measures. D-ID’s facial deidentification technology is specifically designed to make it difficult for AI to overcome and will serve to take individuals and organizations out of harm’s way.”
“More and more data about every one of us is being generated at an alarming pace from a multitude of devices. Keeping our privacy safe from the risk that face recognition brings, requires a combination of decisive regulation and powerful technology. D-ID is one of the leaders on the technology side and their groundbreaking technology helps organizations ensure that the privacy of customers, employees and citizens remains safe.’
Organizations, as well as individuals are becoming more aware of how their information is used – and misused. While user activity has abated on sites like Facebook, people, by nature, are social beasts. Companies and consumers will continue to communicate online, create, upload, share and store images and videos. D-ID is setting out to give them the security and peace of mind in the process.