VGG16-random fourier hybrid model for masked face recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• The COVID-19 pandemic necessitates mask-wearing, significantly impacting facial recognition system accuracy.
• Occlusion of facial features by masks poses challenges for existing recognition technologies.

Purpose of the Study:

• To develop and evaluate a robust method for masked face recognition.
• To address the accuracy degradation of facial recognition systems due to face masks.

Main Methods:

• A novel approach combining a cropping-based strategy focusing on the upper face (forehead and eyes) with an improved VGG-16 architecture.
• Utilizing transfer learning with VGG-16 for extracting discriminative features from un-occluded facial regions.
• Employing Random Fourier Feature extraction to reduce feature vector dimensionality and mitigate bias.

Main Results:

• The proposed method demonstrates superior performance in masked face recognition compared to state-of-the-art techniques.
• Experiments conducted on multiple benchmark datasets (Georgia Tech, HEAD POSE, Robotics Lab) validate the approach's effectiveness.

Conclusions:

• The developed masked face recognition system effectively overcomes challenges posed by facial occlusions.
• This research offers a significant advancement in maintaining the reliability of facial recognition technology in real-world scenarios involving mask usage.