ULN: An efficient face recognition method for person wearing a mask.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Face recognition systems struggle with significant occlusions like masks.
• Mandatory mask-wearing during the COVID-19 pandemic disrupted existing face recognition technologies.
• There is a critical need for robust algorithms capable of handling masked faces.

Purpose of the Study:

• To develop a mask-robust face recognition algorithm.
• To address the challenge of recognizing individuals despite facial occlusions caused by masks.
• To improve the performance of face attendance and surveillance systems in mask-wearing scenarios.

Main Methods:

• Generated synthetic masked facial images to augment training datasets.
• Proposed a novel Upper-Lower Network (ULN) architecture for masked face recognition.
• Utilized a pretrained upper branch for supervisory information and divided high-order features into upper and lower parts.

Main Results:

• The proposed ULN method demonstrated effectiveness in recognizing masked faces.
• The algorithm achieved superior performance compared to existing state-of-the-art face recognition methods.
• Experiments confirmed the robustness of the developed approach against mask occlusion.

Conclusions:

• The Upper-Lower Network (ULN) provides an effective solution for face recognition with masks.
• The method successfully alleviates the data shortage problem and improves recognition accuracy.
• This research contributes a vital tool for reliable identity verification in the era of mandatory mask usage.