Researchers at Spain's Universitat Oberta de Catalunya and Universidad Autonoma de Madrid, along with colleagues at the Massachusetts Institute of Technology, found that facial recognition systems show bias toward certain facial expressions.
The researchers experimented with three leading facial recognition models trained on open source databases like VGGFace2 and MS1M-ArcFace and benchmarked against four corpora. The researchers used Affectiva software to classify images from the benchmark corpora by expression.
They found that "neutral" images surpassed 60% across all datasets, and 90% of images featured "neutral" or "happy" people, with results varying by gender.
Said the researchers, "The lack of diversity in facial expressions in face databases intended for development and evaluation of face recognition systems represents, among other disadvantages, a security vulnerability of the resulting systems."
From Venture Beat
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA