Masked Face Recognition Using Histogram-Based Recurrent Neural Network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• The COVID-19 pandemic highlighted limitations in existing face recognition systems when encountering masked individuals.
• Accurate masked face recognition (MFR) is crucial for security and identification systems.
• Current systems struggle to detect or accurately recognize faces obscured by masks.

Purpose of the Study:

• To develop an effective method for masked face recognition (MFR).
• To address the underfitting problem and reduce computational load in MFR systems.
• To propose a novel approach combining feature extraction and deep learning for enhanced MFR.

Main Methods:

• The proposed method, histogram-based recurrent neural network (HRNN), integrates Histograms of Oriented Gradients (HOG) for feature extraction.
• Recurrent Neural Network (RNN) is employed for the deep learning component.
• The HRNN method was trained and validated on large-scale datasets, including RMFD and LFW-SMFD.

Main Results:

• The HRNN method achieved a high true acceptance rate (TAR) of 99%.
• The approach demonstrated effectiveness in overcoming underfitting issues.
• Reduced computational burdens were observed during large-scale dataset training.

Conclusions:

• The proposed HRNN method offers a viable and high-performing solution for masked face recognition.
• The integration of HOG and RNN effectively enhances MFR accuracy.
• The HRNN method proves effective on benchmark datasets, validating its practical applicability.