Humans Versus Deep Learning: Detection of Face Morphing as a Peril

Abstract

Face recognition has seen a surge in popularity across various sectors. It is now the most commonly used biometric safety system. Even high-profile border security and immigration in various airports are now incorporating this in place of conventional methods, and so it is vital to tackle all its shortcomings. Face morphing is a peril to face recognition, fooling both machine-based systems and humans. Various morphing algorithms are capable of making undesirable morphs that allow forgers to use a single ID for multiple people. This paper performs a perusal of existing works in face morphing detections. It also discusses the results of a conducted survey that accesses the capability of human subjects to recognize morphed images. The performance of VGG19 and GoogLeNet on AMSL Face Morph Image Data Set is also juxtaposed with human recognition intelligence in order to edify how face morphing attacks are a peril to face recognition systems.