Facial Expression based Person Authentication

Abstract

High quality automatic facial expression based person authentication system is practically difficult mainly due to poses in face. This work paves way to develop a more perfect automatic person authentication system using facial expressions. In this work, ways to extract automatically pose free face images from video taken in normal room condition, determining mouth region, extracting features along with performance comparison in person authentication during normal and smile facial expressions is explained. The system contains two stages. In first stage, automatic pose free image selector is used to collect pose free face images from videos of ten persons taken in two sessions each with normal and smile facial expressions with poses. Testing on images taken from forty videos of resolution 640 x 480 the system identified and extracted pose free face images automatically which are 100% perfect pose free face images. The rejected images may have pose free images, but it will not affect the working accuracy of the system even though may reduce its speed, but not significantly. In stage second, automatically selected pose free images of mouth during normal and smile facial expression from the twenty videos of first session is used for training an auto associative neural network. Images from the second session of twenty videos are used to test for person authentication. The results clearly show that normal face gives more performance than smile facial expression for person authentication by accepting authentic persons and rejecting impostors. Equal error rate is used to calculate the performance of the person authentication system. Equal error rate for person authentication using normal face is 0.32% whereas with smile facial expression is 0.4%. The person authentication system is considered more efficient if the equal error rate value is lower.