Busting Movie Myths: Face Recognition, Fingerprint, & Biometric Accuracy
You’ve seen this in movies countless times before: the smoke-filled corridor, a high-tech face scanner sits just outside the bank vault. Our hero walks by and a laser sweeps across his face…the vault opens immediately.
TV and movies always seem to show face recognition, fingerprints, retina scans, and other biometrics to be perfectly accurate. In real life, they can be very accurate, but perfection is a funny thing – no biometric method can technically be “perfect.” So the goal is to be as accurate as possible while making things convenient and quick enough for real people to actually use. And the movies never acknowledge factors that affect accuracy in real life, like the quality of the face/finger/retina scan or the technology used to compare the scans…not to mention lighting issues or smudges on the scanners. In Hollywood, everything just works all the time, every time.
So, here’s the scoop on how biometrics really work and what actually determines real-life accuracy:
Enrollment
The authorized person establishes their identity, usually by taking one or more reference images that are converted into a numerical representation (called a template) and stored in a database.
Authentication
Now that we have a base template enrolled, each time a user authenticates, three steps happen:
Acquisition – a face, fingerprint, or retina is captured as an image.
Extraction - the image is converted into a numerical representation (template).
Comparison – this new template is compared to the existing template from enrollment.
If the match is “close enough” the user is recognized. Those are the basics.
So what about accuracy?
The specifics of how each step is done, what details are examined during extraction, how many points are used, and how closely the templates must match are part of what differentiates one biometric company and solution from another and largely determines performance and accuracy.
With all biometrics, the more details we examine during the extraction step, the less likely a false recognition is. Some solutions consider as few as 35 detail points sufficient. With FastAccess Anywhere Face Recognition, we require at least 400 quality points before converting them into a template.
When a newly created template is compared to the previously enrolled template(s) stored in the database, an exact match is extremely unlikely (remember – lighting, smudges, positioning differences). So virtually all biometric programs use a "strength" number to define how close the match is.
How close is “close enough?”
A very “loose” match (low strength number) can result in you being recognized almost every time but at the expense of accuracy – someone else’s finger or face can be mistaken for yours. A very “tight” match (high strength) can prevent those false recognitions but also makes it less likely that you will match to even your own previous templates. This is why Sensible Vision has a number of patented technologies that allow FastAccess to use a very tight match for security purposes while still providing frequent and quick recognition.
The bottom line
In order for a biometric solution to be accurate, it must not only extract the proper details from the finger or face, but also match them to the enrollment database with an exacting comparison.
Next time we will take a closer look into how Sensible Vision tests its face recognition for both accuracy and performance.
George Brostoff, CEO, Sensible Vision, Inc.