A Survey and Analysis of Face Spoof Detection Techniques based on Image Distortion Analysis Convolution Neural Networks

Abstract

Facial recognition has become a pillar in biometric technology for authentication in finance, surveillance, and consumer applications. Nevertheless, it remains severely susceptible to presentation attacks (PAs), including printed photographs, video replays, and 3D mask attacks. To curb threats, scholars have developed a wide range of face anti-spoofing (FAS) methods, which can be generally classified into texture-, motion-, depth-, and image-quality-based methods. This paper gives an overview of this field, especially Image Distortion Analysis (IDA) and hybrid systems like AdaBoost-based Convolutional Neural Networks (ABCNNs). We examine their advantages, constraints, and applicability at unconstrained conditions with the aid of comparisons with datasets and evaluation procedures. Also, the latest developments, such as domain adoption, meta-learning, transformer-based architecture, and multi-model fusion, are presented. This research is expected to influence the creation of robust, scalable, and explainable face spoof detection systems applicable in the real world by synthesising the information presented by previous research and by pointing out emerging trends.