Masked face recognition with convolutional visual self-attention network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• The COVID-19 pandemic necessitated widespread face mask usage, significantly impairing automated face recognition systems.
• Existing face recognition technologies struggle with the occlusion introduced by masks, necessitating advancements in masked face recognition (MFR).

Purpose of the Study:

• To develop an improved method for masked face recognition (MFR) that overcomes the limitations of current systems.
• To propose a novel deep learning architecture, the Convolutional Visual Self-Attention Network (CVSAN), for enhanced MFR.

Main Methods:

• The proposed Convolutional Visual Self-Attention Network (CVSAN) integrates self-attention mechanisms with convolutional neural networks.
• Self-attention is employed to model long-range dependencies, complementing the local feature extraction of convolutional layers.
• A new large-scale dataset, Masked VGGFace2, was generated using face detection algorithms for training the CVSAN model.

Main Results:

• The CVSAN model demonstrated significantly improved performance in masked face recognition compared to existing algorithms.
• The integration of self-attention effectively addressed the challenges posed by mask occlusion in face recognition tasks.

Conclusions:

• The CVSAN architecture offers a robust solution for masked face recognition (MFR) challenges.
• The developed Masked VGGFace2 dataset provides a valuable resource for training and evaluating MFR systems.