Biometrics make use of our most unique physical features and behaviors to serve as digital identifiers that computers and software can interpret and utilize for identity-related applications. They can be used to identify someone in a biometric database, or to verify the authenticity of a claimed identity.
Facial recognition technology applies the science of biometrics to a user based on his or her facial features. Facial recognition algorithms create a biometric template by detecting and measuring various characteristics or feature points, including location of the eyes, eyebrows, nose, mouth, chin, and ears. Two templates are compared to yield a match score, which indicates the likelihood that the two images belong to the same person. Liveness detection may also be applied to ensure that the source of the biometric sample is not a digital or paper reproduction.
Law enforcement agencies such as the FBI use facial recognition technology to search criminal watch lists and help conclusively identify a person of interest. They can match a recently -taken photo or video against a database of existing photos to help identify that person. Governments around the world use facial for biometric identification for a variety of applications including border security, fraud prevention, and citizen ID.
More recently, mobile authentication has emerged as the most common main private-sector use case for facial biometrics, where our facial images can be used in place of a password. Users can now log onto their devices using facial recognition as a convenient and secure alternative to PINs and passwords. Incorporating facial recognition into a mobile app, such as for banking or other secure applications, allows the app provider to have more control of the security features and user experience, and also achieve more consistency across customer devices.
Facial recognition’s advantages
Each biometric modality offers advantages and disadvantages. Facial recognition offers several advantages over other biometrics:
- Proliferation of digital facial image data.: The billions of existing digital facial images from countless sources are extremely useful for algorithm training purposes, particularly for algorithms using machine learning techniques.
- Enhancement of manual facial recognition.: Biometrics can be used in concert with human visual facial recognition processes, such as comparing a live person to their facial image on their ID card. Facial recognition technology can be used to automate or enhance this process and provides greater matching accuracy. For just about any process where a person’s face is used by a human to verify their identity, biometrics can be used to improve it.
- Ubiquity of cameras on mobile devices.: Nearly all smartphones, and tablets, and laptops have built-in front-facing cameras that enable high-quality “selfie” shots. This makes it convenient to collect a live facial recognition sample for comparison against a template.
- Conducive for use with other modalities for mobile authentication.: Facial image capture using the front-facing camera on a phone can be performed passively and simultaneously during capture of other modalities such as voice and keystroke to improve matching performance and liveness detection.
Facial recognition’s challenges
The following challenges must be addressed for facial recognition to be useful:
- Wide variety of environmental capture conditions (e.g. lighting, shadows) make accurate matching more challenging. Pose variations, aging, glasses, facial expressions, and facial hair can also make matching more difficult. Differences between camera sensors and settings can also have an impact.
- High availability of facial images means fraudsters can more easily obtain images of potential fraud victim that can be used for spoofing.
Overcoming the challenges
Several steps can be taken to optimize security, performance, and user experience.
Automate capture of high-quality images
Software that automates capture of facial images analyzes the streaming video frames in real time. Image capture is automatically triggered once focus, facial positioning, lighting, and other image-capture details are verified for compliance with quality standards.
The image is analyzed again and optimized post-capture. Automatic scaling, rotation, cropping, brightness, and contrast enhancements optimize the quality of an otherwise non-compliant image so the photo doesn’t have to be retaken.
Perform robust liveness detection
Fraudsters may attempt to spoof a facial recognition algorithm with non-live digital videos and images obtained online. Liveness detection algorithms distinguish between a printed, digital, video, and a live facial image.
Passive liveness detection looks for indicators of a non-live image such as inconsistent features between foreground and background.
Active liveness detection prompts the user to blink or shake his or her head to ascertain liveness.
The combination of active and passive methods will yield the highest performance matching. Facial recognition analysis applications might also leverage a second modality such as voice to help assure liveness.
Use high-performance algorithms
In the past few years, machine learning algorithms have become the predominant method for automatically extracting the above information and then comparing it against other images.
Aware’s facial recognition products
Aware products for facial recognition:
- Knomi: Mobile authentication with facial matching and liveness detection
- PreFace: Face image autocapture and processing on mobile device or desktop.
- Nexa|Face: Face matching algorithm SDK.
- FaceWorkbench: Forensic examiner workstation application
- AwareABIS: Automated biometric identification system.
Click here to learn more about Aware’s portfolio of facial recognition products and services.