You may have unlocked your phone today with your face, but do you really understand how the biometric technology known as facial recognition works? In this article you’ll learn how using an individual’s physical characteristics proves a better choice than traditional security methods that are easily lost or stolen, such as personal identification cards, passwords/PINs, and keys.
While one of the most ubiquitous use cases is mobile phones, the possibilities extend far beyond personal mobile devices and include significant benefits in safety, security, and efficiency across a variety of industries. As a compelling, comprehensive, and rewarding biometric technology, it is vital to understand how facial recognition works, what sets it apart from other biometric modalities, how it can be deployed and optimized, technical considerations and specifications, varied use cases, and of course, its potential.
Recent advancements in AI technologies have unlocked a wealth of new use cases for facial recognition. This popular technology identifies facial vectors and features then matches them with pre-enrolled individuals.
AI algorithms run mathematical equations to create a template of an individual by measuring their facial variables. Unique measurements are processed, such as nose depth and width, forehead length, and eye shape. Facial recognition technology then compares this newly generated template with existing templates in a database. If there is a match, it can confirm an individual’s identity.
CyberLink has created an AI-based facial recognition engine called FaceMe®. By building on its facial recognition expertise, CyberLink has leveraged deep learning and neural networks to push the boundaries of these technologies. The result is one of the world’s most accurate, secure, and flexible edge-based solutions.
There are three main biometric technologies:
Use the chart below to understand the strengths and weaknesses of each solution.
Accuracy
Ease-of-Use
Speed
Hygiene
Special Hardware
Hardware Cost
Remote Enrollment
Block listed & Stranger Prevention
Very Good
Very Good
While it can be fast and highly accurate, Iris recognition requires specific, expensive cameras. The downside to fingerprint verification is the hygiene concern associated with touch verification — in addition to dirt or oil from fingers interfering with the efficacy and integrity of these sensors. Facial recognition is recognized as the superior method of vision technology because it is affordable, hygienic, and flexible.
Facial recognition can perform tasks beyond face detection and face recognition – making it the most powerful and relevant AI biometric technology. With a more robust and feature-forward facial recognition platform, such as FaceMe, the benefits only grow while biases decrease.
A facial recognition engine performs several steps, as outlined below:
Face detection is the first step. The technology scans the area for any full or even partial human faces. Fast, precise face detection is necessary before the technology can begin the face match and recognition processes. FaceMe can detect more than one face at a time, and count how many faces are simultaneously present.
Feature extraction is the next step. After a face has been detected, a template is required. The engine extracts an n-dimensional vector set - a template - from the detected facial image. Achieving a precise template requires a high “n” value.
If the goal is to verify a person’s identity, the engine will then perform a face match. A match uses a 1:1 method. The newly extracted facial template is either matched with a facial template in the pre-enrolled database, or cross-referenced with a piece of identification to perform a match. Some of the most common uses of face match occur on mobile devices, such as Apple’s Face ID which unlocks your phone, or mobile banking where you log in to your financial service portal with your face.
In this function the question is no longer, “Is this person who they say they are?” but “Who is this person?” FaceMe can answer this question using a face search. This time the engine completes a 1:N search, comparing an individual’s template against pre-enrolled faces in the database to find the best match and confirm the person’s identity. To ensure privacy, no actual images of faces are stored on the platform — FaceMe only stores encrypted data. The most common use of face search is a security and surveillance camera system to verify that people on a premises belong to a particular company. If their face was pre-enrolled in the company’s facial database, face search will confirm it is them and grant them access.
Some key use cases require additional features, such as the following:
This is an important feature in smart retail and digital signage which allows for customized ads and messaging while collecting detailed visitor statistics. Face attribute detection is a facial analysis that identifies and analyzes characteristics such as age, gender, mood, and head orientation or movements (e.g., nodding, shaking).
As a result of the Covid-19 pandemic, mask wearing has become a significant factor for ensuring health and safety. Mask detection is one of the newest and most valuable features in facial recognition technology. CyberLink’ FaceMe SDK is a cross-platform facial recognition SDK (Software Development Kit) that offers optimized mask detection. It can even perform facial recognition when the individual is wearing a mask. The software recognizes health-compliant masks and verifies if the nose and mouth are properly covered, all while performing highly accurate face detection and recognition. Additionally, CyberLink’s FaceMe Security allows existing security solutions to upgrade their mask detection features so that employees can enter their offices without having to remove their masks.
Biometric fraud occurs when a person uses someone else’s photo or video in front of a camera to gain access. Anti-spoofing technology provides protection against these malicious actions through secure and accurate liveness-detection.
2D cameras, such as USB webcams, catch fraud through interactive and non-interactive measures. Interactive measures detect natural head or facial movements to confirm the presence of a live person. Non-interactive measures are unique to each solution provider’s AI algorithm.
3D cameras allow for near instantaneous anti-spoofing by performing depth detection without the need for additional interactive measures. 3D cameras are costlier but generally provide a superior experience. 2D alternatives, however, can provide accurate anti-spoofing at a fraction of the cost. A few of the 3D camera options compatible with FaceMe are Intel RealSense, 3D cameras on iPads and iPhones, Orbbec, Himax, Altek, and eYs3D. FaceMe supports both 2D and 3D cameras.
CyberLink’s FaceMe facial recognition technology passed iBeta’s spoofing attack detection test by detecting and rejecting all impersonation and substitution attempts using the test’s photos and videos.
Accuracy in facial recognition is measured in terms of false matches. The most precise facial recognition engines are characterized by a low false non-match rate (FNMR) and an extremely low false match rate (FMR). A false non-match is the failure to match two facial templates from the same person, whereas a false match is when a person’s face is matched with someone else’s.
The governing body that measures the accuracy of facial recognition algorithms is The National Institute of Standards and Technology (NIST). NIST’s Facial Recognition Vendor Test (FRVT) evaluates algorithm performance using different datasets. For example, the VISA category tests the algorithm’s ability to correctly identify an individual based on a passport photo, while the WILD test uses random, non-constrained photojournalism-style images.
FaceMe achieved 99.48% at 1E-6 accuracy in the NIST VISA-Border test — one of the highest ratings. 1E-6 means the False Match Rate (FMR) is 1 in 1,000,000 (0.000001). FaceMe also demonstrated top-tier accuracy for WILD (97.00% at 1E-5). By comparison, a smartphone with Face ID offers about 96% at 1E-4 accuracy.
Beyond the specific algorithm, some of the factors that affect accuracy are camera resolution, camera positioning, lighting, cleanliness, and camera type. Facial recognition engines work adequately with 720p cameras, but a 1080p resolution is generally recommended.
The best facial recognition solution for you depends on your specific needs. Are you a developer who just needs a facial recognition SDK or API to start developing your own applications? Or are you a business owner or decision-maker that wants an out-of-the-box facial recognition solution?
FaceMe has two different types of solutions for developers, depending on the development and coding requirements of their facial recognition application.
Facial recognition is the future of AI biometric technology. The convenience and accuracy it provides are already being realized to streamline processes and customer services. Businesses are keeping their employees safe by automating secure access control in the office. Retailers are enhancing customer experiences in their stores. Manufacturers are simplifying access to restricted areas. Banks and fintech companies are introducing much stronger authentication and cutting-edge security controls. And that's just the beginning.
Without a doubt, the ethical deployment of facial recognition technology is on track to make the world a better place.
ContactOur team of experts will be happy to answer your questions and schedule a demo. Free evaluation versions of the FaceMe® are available to qualified contacts.
