Using face quality ratings to improve real-time face recognition

Abstract

A Face Quality Rating (FQR) is a value derived from a face image that indicates the probability that the face image will be successfully recognized by a specific face recognition method. The FQR can be used as a pre-filter in real-time environments where thousands of face images can be captured every second by multiple surveillance cameras. With so many captured face images, face recognition methods need to strategically decide which face images to attempt recognition on, as it is prohibitively difficult to attempt recognition on all of the images. The FQR pre-filter optimizes processor time utilization resulting in more people being recognized (faster and more accurately) before they leave the surveillance cameras' views. We generate FQR values using Multiple Layered Perceptron (MLP) neural networks. We then use these MLPs in a real-time environment to experimentally prove that FQR pre-filtering improves the speed and accuracy of any real-time face recognition method... © 2009 Springer Berlin Heidelberg.