Onfido working to eliminate demographic differences in biometric facial recognition technology
The global biometric facial recognition market is estimated to reach a CAGR of 14.41 percent by 2028, with growth generated by a high interest in surveillance systems to boost security, tech innovation in recognition deployments and facial recognition integration with consumer electronics, according to the Global Facial Recognition Market Forecast 2019-2028 released by ReportLinker.
Against the backdrop of this opportunity, biometrics company Onfido has been actively working on improving its machine learning technology to reduce facial recognition bias associated with dark skin tones, writes Axios. This is an important step in identifying all races and ethnicities consistently and accurately, because the COVID-19 pandemic has driven a higher need for contactless, remote identity verification technology.
Onfido’s technology analyzes facial images on official identity documents by cross-comparison with a selfie or video submitted by the user. The algorithm can establish is the person in the document and in the selfie are one and the same.
The company has recently been awarded a £5 million (US$6.6 million) grant to invest in the development of biometric fraud detection, data anonymization and document fraud prevention technologies for small and medium enterprises (SMEs).
“Essentially, we’re replicating what happens in-person in a bank branch and making it digital,” Husayn Kassai, Onfido’s CEO, told the publication.
The company’s director of product Susana Lopes told Axios that they “changed the way it trained the algorithm to help it learn from an unbalanced dataset.”
Onfido’s face matching False Acceptance Rate (FAR) is now 0.01 percent, at a False Rejection Rate of 1 percent, the highest ever reported accuracy for comparing faces on identity documents with selfies or videos. The company has reported 10x overall improvement and a 60x false acceptance improvement for faces of African descent.
Onfido is part of the U.K.’s first Information Commissioner’s Office (ICO) data protection and privacy sandbox to address algorithmic bias in machine learning models. The company’s focus is on racial bias for biometric-based identity verification.