Making interruptions on video calls disappear with biometric recognition

A rudimentary way of removing specific uninvited people from a video conference using facial recognition algorithms has been demonstrated by researchers in Romania.

The tool is called Video Stream Privacy and is designed to protect the privacy of people who are caught, unwanted, on a web camera, according to a pair of researchers at Alexandru Ioan Cuza University.

Their brief paper is short on some details but clearly describes the process.

A so-called anchor’s face is scanned to differentiate it from that of unwanted participants. A generic background image is inserted behind the anchor. Any additional face is scanned and analyzed by a facial recognition algorithm (Facebook’s DeepFace library) and removed from the frame just as the real background is replaced by the background.

Beside DeepFace, the tool uses Bodypix body-segmentation and YOLO object-detection algorithms.

The researchers write that they were able to remove a person’s face even when the anchor occupies the same area of localization as the person not wanting to be recorded.

In their experiments, they found that processing every frame caused laggy streaming, which could be addressed by processing every 60 frames. A future change might be using a few frames of the actual background to hide unwanted faces rather than a virtual background.

The research is designed to protect the privacy of people caught on a web cam inadvertently, but there is nothing in the report about protecting the data privacy of people whose faces are scanned and analyzed as part of the process.

It also is not clear how it would work when someone passes behind the anchor. Motion and a profile image might be problematic.

A team of researchers pointed out in 2020 that video conferences can be easily infiltrated, and participants identified with biometrics.

Article Topics

biometric identification  |  biometrics  |  biometrics research  |  data privacy  |  video conferencing