2 min readAs biometric identification applications, such as facial recognition, become more common in everyday life, there's a growing demand to make them part of enterprise solutions. It is worth finding out just how effective biometric identification is.
Biometric identification has been around ever since fingerprint scanning technologies were introduced in 2013. While biometric identification can be fast, effective and secure, the technology does have certain limitations. One issue revolves around physical characteristics not being really private and, therefore, key attract criminal activity because of the loopholes present within the technology.
Why facial recognition?
It’s always best to add an extra security layer to authentication, especially when your personal information is at risk of being publicly visible.
How about the depth camera?
Fraud detection is one of the main concerns to take into account when implementing biometric identification. Modern facial recognition algorithms can easily be infiltrated when presented with a human’s picture or video as an input (technology is not 100% accurate, yet). Sadly, there is no “only software” solution to make these systems work only on recognizing a real human face instead of a photo. A depth camera can add the required dimension to increase recognition accuracy and detect possible fraud. Usually, these cameras can capture a 3D image by measuring the distance between self-projected infrared dots invisible to the naked eye. Modern consumer-available depth cameras can be efficient at a range between 0.3-10m, with the highest precision near 1mm.
SoftServe’s approach
SoftServe uses raw camera images (RGB) to detect a human face with computer vision. Areas on the human face are used to detect specific facial features—these are helpful to locate specific regions of interest (i.e. nose, forehead, lips, eyes, and more). The final stage is to get ROI’s distance from depth stream and validate on SoftServe’s pre-defined set of rules to distinguish whether it’s a real human face or just a picture.
Caption: Real face
Caption: Photo of a face
Ways to improve
Thus, SoftServe’s approach can filter security loopholes or threats by implementing users’ photos or videos. Facial recognition cannot accurately detect an actual 3D-mask or puppet, and therefore SoftServe are working on implementing sequential recognition of multiple face frames obtained via a depth camera to resolve such an issue.
Conclusions
Though biometric identification, especially a facial recognition system, is not entirely secure, depth cameras can assist to make it secure. In relation to other biometric recognition techniques (i.e. fingerprint, iris, and similar) facial recognition can be a more reliable and convenient way of authentication for fraud detection.