A Productive On-device Face Authentication Architecture for Embedded Systems

Abstract

AbstractFace authentication is one of the effective cyber security confirmation techniques used nowadays, and it is designed and implemented using embedded systems. Face authentication systems need cloud or server-based platforms to manage the face database and to perform image processing and deep learning. The world is moving towards a situation where data privacy has a severe concern compared to any other era. Sharing private data to remote servers for face authentication is highly sensitive. The existing works focus on machine learning and deep learning aspects rather than the practical or implementation aspects of embedded systems. In this paper, we focus the major benchmark inventions on the embedded implementation perspective. We have practically deployed and validated the existing architecture on various embedded platforms such as smartphones, development kits and tablets etc. The architecture was evaluated on different versions of the Android operating system as well. We have analysed and compared different practical aspects such as model size, latency, power consumption, efficiency, and memory usage, along with the machine learning aspects such as prediction accuracy, precision etc.KeywordsOn-deviceFace authenticationEmbedded systemsContactless authenticationMasked face recognition