Improving Face Matching Accuracy

The aim of the project is to explore how human operators making face matching decisions can be optimally assisted by state of the art face matching algorithms, ultimately helping people to make quicker, more effective and more accurate decisions. These explorations will be used to design an integrated human-machine team for face matching that optimises accuracy in face matching tasks and will have realistic applications in settings such as in the security and border controls therefore has significant importance. Face matching is a surprisingly error prone task with substantial variation in individual human performance. Numerous factors such as age, gender and race along with facial expressions and emotions can influence the accuracy of facial recognition. However, face matching by human observers is widely used to verify identity in applied, professional settings, where the ramifications of an error can be severe and potentially life changing. Recently there have been major gains in the accuracy of automated facial recognition algorithms using machine learning and artificial intelligence. The project therefore aims to bridge the gap between the two fields and help humans make the most of technology. Laboratory research using constrained scenarios have demonstrated that fusing algorithm scores with the ratings of top human performance can provide almost perfect accuracy on a challenging face matching task. To obtain such high levels of performance in more natural scenarios will require research to understand how best to form human-machine teams for this task. We propose a series of experiments that will first use established tools to obtain human and machine baseline results. From this we will investigate novel ways to combine human and machine intelligence to improve facial matching decisions. Of particular interest will be investigating the concept of calibrating trust between human and machine to facilitate optimal human-machine team performance. The findings from these first two stages of research can subsequently be used to inform best practice in designing processes that minimise the risk of errors and misidentification in applied settings.