Parametrically controlled synthetic imagery experiment for face recognition testing

Abstract

The use of synthetic imagery for testing biometric systems is relatively new and in need of further exploration. In this paper, we describe methods and procedures for using synthetic images generated from shape and texture data to refine and extend the current state of the art of face recognition performance testing. Two example experiments are presented based on the canonical Facial Recognition Vendor Test 2000---pose experiments and temporal experiments. We demonstrate how the use of synthetically generated face models (and resulting images) can enhance and extend existing test protocols and analysis. These methods and results will be of use to developers and practitioners alike.