Why Your Authentication Must Include Presentation Attack Detection
Facial recognition is fast becoming the preferred method of biometric authentication due to its non-intrusive and contactless nature. In fact, it is second only to fingerprints in adopted use worldwide. According to a recent FICO survey, 65 percent of Americans and 64 percent of Canadians are willing to provide biometric information, such as facial images, to their banks. However, facial recognition is particularly vulnerable to presentation attacks in unsupervised settings, such as in mobile banking and remote employee onboarding. Because presentation attacks are performed at the sensor level, fraudulent actors do not need access to the backend of the system to compromise the integrity.
Presentation attacks include 2D methods, such as using a printed photo or replaying a video of the subject, and more sophisticated 3D attacks, involving custom silicone masks. There are a number of methods available on the market to address these presentations attacks, but most of them fail to detect 3D attacks. Additionally, as the tools attackers use evolve, such as more advanced cameras, display devices, and manufacturing methods, the ability to detect these presentation attacks becomes more difficult in real-world settings.
Due to society’s growing reliance on using biometrics as a method of securely accessing private and valuable information, any secure face biometric authentication solution must include reliable and robust 3D PAD anti-spoofing. Without PAD measures, even the most state-of-the-art facial biometric systems are vulnerable to simple attacks. Face authentication with active and passive liveness detection and anti-spoofing technology provides enhanced trust by ensuring a live face is presented and is attached to the in-the-flesh authentic individual.
Attack Detection Through AI
ImageWare’s Biointellic solution easily detects and stops both 2D and 3D attacks by breaking presented images up into pieces and analyzing each one with a sequence of Deep Neural Networks (DNN). These DNNs search for artifacts to determine if an image is spoofed, such as skin texture, lighting, foreground and background irregularities, edge and encoding artifacts, and comparative depth. Using the results of these DNNs, Biointellic generates a score to determine if the image is live or spoofed.
The DNNs, created using machine learning techniques, ingest large amounts of labeled samples to learn what makes up a real or spoofed image. After analyzing the labeled samples, the DNNs then analyzes unlabeled samples and evaluates the error rates. The process is repeated, tuning the algorithm, until the error rates are significantly minimized.
iBeta, a software testing and quality assurance company for the world’s most trustedand brands, tested Biointellic for compliance with ISO/IEC 30107-3 level 1 and achieved an Attack Presentation Classification Error Rate (APCER) of 0%. One-hundred and eighty presentation attacks were attempted over an eight-hour testing period, and Biointellic successfully identified every one of them.
ImageWare’s Biointellic is the world’s first zero-friction presentation attack detection solution for use with mobile devices. Our solution uses standard mobile device cameras and a single photo for analysis. Biointellic does not require any special movements or environmental conditions to determine a presentation attack. Instead, the Biointellic decision engine breaks the image into small regions and compares adjacent regions for discrepancies. ImageWare’s Biointellic can determine presentation attacks such as high-quality latex masks, photos, digital imposition, and video playback.
This level of security and convenience is available now - schedule a demo by clicking the button below to see our proofing in action today: