A Systematic Literature Review on Face Morphing Attack Detection (MAD)

Abstract

Morphing attacks involve generating a single artificial facial photograph that represents two distinct qualities and utilizing it as a reference photograph on a document. The high quality of the morph raises the question of how vulnerable facial recognition systems are to morph attacks. Morphing Attack Detection (MAD) systems have aroused a lot of interest in recent years, owing to the freely available digital alteration tools that criminals can employ to perform face morphing attacks. There is, however, little research that critically reviews the methodology and performance metrics used to evaluate MAD systems. The goal of this study is to find MAD methodologies, feature extraction techniques, and performance assessment metrics that can help MAD systems become more robust. To fulfill this study’s goal, a Systematic Literature Review was done. A manual search of 9 well-known databases yielded 2089 papers. Based on the study topic, 33 primary studies were eventually considered.Anovel taxonomyofthe strategies utilized inMADfor feature extraction is one of the research’s contributions. The study also discovered that (1) single and differential image-based approaches are the commonly used approaches for MAD; (2) texture and keypoint feature extraction methods are more widely used than other feature extraction techniques; and (3) Bona-fide Presentation Classification Error Rate and Attack Presentation Classification Error Rate are the commonly used performance metrics for evaluating MAD systems. This paper addresses open issues and includes additional pertinent information on MAD, making it a valuable resource for researchers developing and evaluating MAD systems.