By: Reza Rassool
Last month, I had the opportunity to discuss facial recognition with Steve Wong at the NAB Show in Las Vegas. As is natural to do with any emerging technology, Steve brought up some common concerns around facial recognition. I’m always happy to answer questions like Steve’s because I believe open and honest discussions about privacy and bias will lead to greater consumer comfortability with this technology, and foster collaboration between facial recognition companies and regulators to ensure these concerns are addressed.
Privacy and Positive Use Cases
RealNetworks takes privacy seriously, employing Privacy by Design Principles, data security measures, and data management tools. SAFR gives subjects the opportunity to opt-in to facial recognition and once in the database, all user data is protected with AES-256 encryption. When these basic privacy concerns are adequately addressed, so many positive use cases are unlocked. Here, Steve and I discuss some of those use cases and how we make subjects active participants in — and beneficiaries of — facial recognition technology:
Clip has been edited for length and clarity.
Reducing Bias in Facial Recognition
Bias-driven false matches are another top concern, and not all facial recognition algorithms are created equal. I’m tremendously gratified that NIST testing found SAFR to be one of the least biased out of more than 100 algorithms, measured by the consistency of match-accuracy across a range of skin tones. That consistent performance was not a coincidence. In this clip I explain how we intentionally trained the SAFR algorithm in a way that minimizes bias:
Clip has been edited for length and clarity.
As facial recognition becomes more widespread, it’s critically important that we address public concerns head on — working to mitigate the potential for privacy infringement and bias at every stage of development, design, and implementation.
Thanks to Steve and the NAB Show for hosting this discussion.
—
Reza Rassool is the CTO of RealNetworks based in Seattle, WA.