Face recognition & Voice authentication – Multimodal biometrics

BioID’s technology is called multimodal because it simultaneously combines multiple biometric traits in a single, simple process. You simply look at the camera and say your name; the biometric authentication software captures and analyzes your face and your voice at the same time.

This approach has several advantages:

• Accuracy: More accurate than any single biometric trait.
• Fraud resistance: It’s harder to fake multiple traits.
• Flexibility: If one trait temporarily changes, for instance your voice when you have a cold, the other trait compensates nicely.

Face and voice recognition

BioID’s highly accurate face recognition starts with our industry-leading face finder

• Far outperforms OpenCV & others at identifying faces & rejecting non-faces
• Faster performance
• Much better user experience - fewer cases of “face not found”

Extremely precisely face positioning: Even a 1% error in eye position can mean 10% reduction in recognition accuracy; precise positioning is one of many reasons our face recognition is so accurate & reliable.

Highly trusted voice recognition: used to testify in court. Voice can be used by itself (e.g. to identify users over the phone) or combined with face recognition in a multimodal scenario.

A major strength of biometric traits for authentication is that they can’t be stolen. However, they can be copied. A poorly designed face detection system could be fooled with a photo or video of the legitimate user, and similarly voice recognition could be fooled with a recording.

A well-designed biometric system requires what is known as “liveness detection”, to ensure that the sample is that of a live user rather than some sort of copy.
BioID’s patent pending “fake defender” liveness detection uses several techniques, including:

• Capturing multiple face images and checking for changes and natural motion. A 3D face moves differently from a 2D photo, and our sophisticated algorithms can detect this difference.
• Guiding the user to turn their head a certain way and verifying that the head was turned in the specified direction.
• Requiring the user to speak multiple words or digits presented in a random order.

BioID’s fake defender can tell the difference between a real live person and an impostor presenting a photo or video to the camera.

BioID’s technology was designed to protect the privacy of user information. End users are rightfully cautious about the use of their biometric information, so we’ve designed our services to keep this information and the user’s identity secure.

Anonymous biometrics: Biometric data is

• Kept completely separate from any other personally identifiable information
• Delivered & stored using an anonymous ID generated by BioID: ensures that the identifier itself contains no personally identifying information (e.g. user ID)
• We never receive any personally identifiable information other than biometrics: we have no idea who users are

With anonymous biometrics BioID separates biometric data from other personally identifying information. Raw photos are converted to an anonymous template, from which the original photo cannot be reconstructed.

Irreversible, revocable template:

• We store no photos, only biometric template, a mathematical representation of biometric data (service provider may store photos, e.g. for audit purposes)
• Photo can’t be reverse engineered from template.
• If user’s template is compromised, service provider can simply delete user & re-enroll under a new anonymous ID

Secure storage: Some biometrics vendors today opt for local storage of biometric data, so that nothing leaves the user’s device. This appears to put the user in control of their data.

However, by putting everything on the client device, the application to be secured is more vulnerable to attack. BioID’s services instead place the anonymous biometric data in a highly secure data center.