Remember meForgot password?
    Log in with Twitter

article imageQ&A: Using AI to fool facial recognition systems Special

By Tim Sandle     Nov 22, 2019 in Technology
As more organizations use facial recognition technology, concerns continue to persist around privacy, anonymity and regulation. For some, 'de-identification' is the answer, as Gil Perry, the CEO/co-founder of D-ID discusses.
D-ID's experts in deep learning, machine vision and image processing have developed two products that can anonymize images and videos of people, which can't be reverse engineered to extract the original facial image. What is interesting is that while a person could identify the face as being of the same person, there are slight tweaks being made to ensure it can't be identified by computer vision software.
The co-founders of D-ID came up with the idea for the company while in the Israeli Defense Forces, recognizing the risks that come from the growing global use of facial recognition technologies. To learn more, Digital Journal cauhgt up with Gil Perry, the CEO/co-founder of D-ID.
Digital Journal: How sophisticated is facial recognition technology?
Gil Perry: At D-ID, we believe that the discussion isn’t about how sophisticated facial recognition technology is, but rather, how common and accurate the systems have become. Today every one of us can lay our hands on simple facial recognition software - or purchase a more powerful one for a fistful of dollars. At the same time, studies show that between 2014 and 2018, facial recognition technology has become 20 times more accurate. What this essentially means is that photos - especially photos that contain human faces - have become super sensitive personally identifiable information. So we should think hard t every time we upload a photo to the internet, because in doing so, we expose it to facial recognition AI that learns our facial pattern - and consequently recognize us.
We’ve seen incidents where a facial recognition engine was able to determine a person’s identity from a single frame of a video feed that showed only half the face. Such examples underline the notion that it’s increasingly important to implement techniques that enhance privacy and ensure regulatory compliance, by removing unnecessary sensitive and biometric data from facial images and videos.
DJ: For what types of use is the technology being applied?
Perry: Facial recognition has some very important use cases, like spotting “the bad guys” in a crowd (think airport security) or allowing people to access their devices, bank accounts or workplace more conveniently through facial biometric authentication. These are all legitimate uses, however, facial recognition can also be used to spy on specific people based on a certain profile (age, ethnicity, etc.). In some countries, facial recognition is used to collect information about citizens. It can also be a powerful tool at the hands of hackers to identify people and track them.
DJ: What are the privacy issues associated with the technology?
Perry: Essentially, our faces are our most sensitive identifiers - even more sensitive than our fingerprints. If facial recognition is applied without supervision, it basically means the end of privacy. Just imagine a world where anyone on the street can know your name, where you live, where you work, the name of your spouse, who your children are and other personal information - all by just looking at your face. We have not yet reached this dystopian world, but without keeping facial recognition in check, that is where we’re heading.
DJ: What’s de-identification?
Perry: De-identification is a way to remove or mask a person’s identifiable information. At D-ID we provide the ability to present and share facial images online in a way that humans will recognize them, but facial recognition engines would not. We create “the same image without the faceprint”. A face in a D-ID photo looks similar to the human eye, but AI-powered facial recognition tools cannot identify the face in the picture. Thus, we protect from misuse of facial data.
DJ: How does the D-ID solution work?
Perry: 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.
DJ: Why did you develop your solution?
Perry:My co-founders and I all served on classified units in the army. At the time we weren’t allowed to upload photos online freely. Back then, we already understood what governments were doing with photo technologies. The reality of the global influx of image sharing, with cameras everywhere, and with the promised excitement of devices like Google Glass at the time – all combined with emerging face recognition technology - gave us a glimpse into the future and how it was moving towards an increasingly dangerous path. That’s when we came up with the idea to build a solution that would allow people to upload photos and use them online, while keeping the biometric data unrecognizable.
DJ: How have you tested out your technology?
Perry:Our solution is already in use with a number of customers around the world, including with Fortune 500 companies and other global organizations. As part of our evaluation process with prospective customers, we successfully test our solution against some of the most advanced facial recognition engines on the market, including some that serve government agencies.
DJ: How will your technology be used in practice?
Perry:Actually, it’s already in use. Our solution will also serve to protect images that are being collected in remote onboarding processes that require the users to scan and upload a facial image or scan a picture ID. In addition, we are in advanced discussions with some of the biggest device manufacturers to embed D-ID on every device (smartphone, tablet, PC, digital cameras etc.) and allow users to add a deidentification feature to their photos as they’re being taken.
More about Facial recognition, Artificial intelligence, machine learning, Data privacy
Latest News
Top News