Facial recognition sees daily press coverage for its capabilities, limitations, and policy implications. It has undergone a revolution since 2012, as the developer community has shifted toward the use of various convolutional neural network architectures. This adoption has led to massive gains in accuracy due to increased tolerance of poor image quality and of aging effects, enabling use of the technology in a larger array of applications, including one-to-one authentication and for one-to-many search. However, technical challenges continue as the technology remains largely proprietary with a vast range in performance across the industry. There remains a need for independent quantitative assessment of the underlying algorithms and the systems for which they are used. The U.S. Department of Commerce’s National Institute of Standards and Technology (NIST) has tracked face recognition capability since the late 1990s and publishes regular reports that constitute the single largest resource for a global audience seeking to understand performance.
In this webinar, Patrick Grother will describe the fundamentals of face recognition, its evaluation, and the performance results from the four ongoing tracks of the Face Recognition Vendor Test (FRVT). The session will detail gains from the new generation of algorithms, failure modes, quantification of demographic effects, aging, scalability, and give an overview of limitations and open research topics. Additionally, the session will cover standards for performance and attack detection measurement, face image quality assessment, face-aware capture, and future activities under FRVT.
Download Associated files: 2020-09-24-hdiac-webinar-facial-recognition-performance-measurement-grother