Challenges for automated face recognition systems

Abstract

Face recognition, as a process of the human visual system, analyses facial properties and contextual information such as body shape. Automated recognition replicates the human process and analyses a face image, which is typically acquired with a visible spectrum sensor. When dealing with automated operational systems, the quality of the captured face image is relevant as it affects the recognition accuracy. Thus, it is necessary to measure the utility of a face sample with both a quality score and complementary measures that can provide actionable feedback. This Perspective addresses challenges and discusses solutions for the optimization of biometric recognition systems specifically related to face image analysis. One of these challenges is the vulnerability to presentation attacks. Consequently, for reliable recognition in non-supervised environments, robust presentation attack detection is required. Moreover, biometric templates must be protected. Finally, acceptability of biometric systems requires fairness of the biometric algorithms and artificial neural networks used. Automated face recognition systems are widely adopted in different operational systems, ranging from authentication with smart personal devices to access control and forensics. This Perspective analyses the critical challenges and proposed solutions for the optimized use of these recognition systems.