Innovatrics Uses Filters to Improve Accuracy of Liveness Detection System

Innovatrics has shared more details about the steps that it has taken to improve its passive liveness detection algorithm. To that end, the company noted that most spoofs are attempted with either low quality images, or (somewhat ironically) with extremely high-quality images that are sharper than they would be if taken in real time.

As a result, Innovatrics was able to improve the accuracy of its system by adding a pair of filters that automatically reject liveness photos that fall above or below a certain quality threshold. In the case of low-quality photos, the system is unable to determine whether or not it is looking at a real face, or the face of an imposter, a doll, or someone wearing a mask. In those situations, Innovatrics’ platform will not accept the photo for the liveness check, and will instead ask the user to submit another image.

Thankfully, the passive liveness system can also provide users with feedback to make it easier to capture a satisfactory photo. For example, those using Innovatrics’ Digital Onboarding Toolkit (which features the liveness tech) may be asked to move to better lighting, and the platform has an auto-capture feature that can automatically take a selfie when the necessary quality conditions are met.

Images that are too crisp, on the other hand, are only as sharp as they are because someone has taken a picture of a photo displayed on a screen. Filtering for sharpness therefore yields similar benefits to filtering for a lack of clarity.

According to Innovatrics, the idea for quality filters during the liveness check emerged from a study that used the same technique to guard against various forms of racial and gender bias in facial recognition systems. Getting rid of poor images makes the entire system more accurate, since it is not trying to make a decision based on edge cases, and is instead only verifying someone’s presence (or their identity) when it is confident in the result.

Innovatrics’ passive liveness system had already achieved iBeta Level 2 certification before introducing the new filters. The company also added new algorithms to the passive liveness solution in October. The Digital Onboarding Toolkit, meanwhile, uses a combination of face and document recognition to verify the identities of people signing up for services online.

–

December 15, 2021 – by Eric Weiss