A survey on face presentation attack detection mechanisms: hitherto and future perspectives

Area of Science:

• Computer Vision
• Biometrics
• Artificial Intelligence

Background:

• Face recognition (FR) systems are advancing for secure authentication.
• Security threats, particularly spoofing attacks, are a growing concern in FR systems.
• Presentation attack detection (PAD) is crucial for distinguishing genuine faces from artifacts.

Purpose of the Study:

• To systematically review state-of-the-art face anti-spoofing techniques.
• To provide insights into face PAD mechanisms, artifacts, datasets, and protocols.
• To identify future research directions in face PAD.

Main Methods:

• Review of computational approaches for face PAD, including hardware-based, handcrafted features, and deep learning methods.
• Analytical overview of face artifacts, performance protocols (HTER, ACA), and benchmark datasets.
• Analysis of twelve recent state-of-the-art face PAD techniques on the REPLAY-ATTACK dataset.

Main Results:

• Current face PAD mechanisms show potential but face crucial issues.
• Key challenges include limited generalization to unknown attacks and inadequate datasets for deep learning.
• Handcrafted features show limited discrimination, and efficient deep learning PAD with smaller datasets is needed.

Conclusions:

• Despite progress, face PAD requires further research to address generalization, data limitations, and efficiency.
• The COVID-19 pandemic presents additional challenges for face recognition and PAD methods.
• This review aims to serve as a reference and guide future research in face anti-spoofing.