New Incode age estimation tool processes biometrics on-device

Prompted by the introduction of online safety regulations globally, Incode has released an on-device age estimation product. An announcement from the company says the tool performs all facial analysis on the user’s device, with “no biometric data, raw or processed, sent to Incode, a client server, or anywhere else.”

The primary selling point of facial age estimation (FAE) is that it does not, at any point, require a user to share a date of birth (as age verification does). Nor does it collect and store any biometric data. According to Chris Allgrove, CTO of Ingenium Biometrics Laboratory, biometric FAE tends to use deep learning algorithms to assess visual ageing indicators like skin texture, face shape and structural ratios. It is a real-time biometric technology that only angages with the user when it is analyzing their face to estimate age.

As such, it has proven popular as an age assurance method for lower risk scenarios, or as an initial check in a step-up age assurance stack. However, it typically comes with a confidence range, meaning the tech needs to be configured to factor in buffer ages.

Three-layered process includes metadata analysis

Incode’s tool provides a binary result: “pass or fail, based on the platform’s configured age threshold.” But the age assurance process itself has three layers.

Incode’s FAE is a “compact model that preserves the accuracy of the server baseline without biometric transmission.” The company says that in testing, it has achieved as low as 1.08 years MAE (mean absolute error, a continuous measure of estimation error) for the strictest age groups.

Security concerns necessitate liveness detection and injection attack detection (IAD); per the company’s post, “keeping the face on-device doesn’t eliminate the need to verify that a real, live person is present.” But for IAD, the company analyzes metadata on its server – meaning that, while all biometric processing is done on-device, some data still runs server-side.

Finally, the system cascades as needed. “When a use case calls for a higher assurance level, due to risk profile, regulatory requirement, or borderline confidence near an age threshold, the system automatically steps the user up to other Incode age check methods, or any other route the customer decides, without interrupting the user experience,” Incode says.

For users, the process just looks like taking a selfie, with automatic capture once optimal positioning and lighting conditions are achieved.

In unstable trust environment, even true claims can be volatile

Incode’s pitch stresses the urgency of finding trusted age assurance methods as regulations take hold in the UK, EU, Australia, and increasingly in the U.S. “More than 32 age assurance laws are in force globally, with requirements expanding into social media, online gaming, and AI platforms. Facial age estimation has emerged as one of the most preferred compliance mechanisms: no government ID required, no external databases, and the only viable option for users who don’t have traditional identification.”

It also names “negative attitudes toward biometric collection” as a driver of the FAE market.

In doing so, the pitch also nicely illustrates the delicacy of selling a biometric product to an ever more privacy-conscious public. The tagline is that Incode’s FAE does all biometric processing on-device. Server-side processing only addresses metadata – but that technicality may not matter to those seeking an opportunity to paint biometrics providers as secret hoarders of user data.

There is no lack of transparency in Incode’s product announcement: it clearly states that “we analyze metadata on our server to prevent injection attacks and tampering of the results.”

Rather, the issue is in the tension between promoting the product and full transparency. Claims that AI software tools are completely privacy preserving inevitably invite challenges. While these may be spurious – see the recent case of Yoti defending its integrity against U.S. researchers – they can do reputational damage, and fast.

There is no obvious answer to the question of how to weigh promotional zeal with the need to build a stable public trust ecosystem in biometric tools in general. But in the biometrics industry, where trust is the most valuable currency, it matters more than in most.

Article Topics

age verification  |  biometrics  |  face biometrics  |  facial age estimation (FAE)  |  Incode