Multi-angle head pose classification with masks based on color texture analysis and stack generalization.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Head pose classification is vital for face recognition but is hindered by mask-wearing due to the COVID-19 pandemic.
• Masks obscure facial features, significantly impacting the performance of traditional head pose estimation methods.

Purpose of the Study:

• To develop an effective head pose classification method for individuals wearing masks.
• To improve the robustness of face recognition systems in real-world scenarios with facial coverings.

Main Methods:

• Utilized HSV color space to extract the H-channel image for relevant information.
• Employed line portrait technique to capture facial contour lines.
• Trained convolutional neural networks on grayscale images for feature extraction.
• Applied stacked generalization to fuse outputs from three classifiers for final classification.

Main Results:

• Achieved high classification accuracies on the MAFA dataset: 94.14% (front), 86.58% (more side), and 90.93% (side).
• Demonstrated superior performance compared to current advanced algorithms for masked head pose classification.

Conclusions:

• The proposed method effectively classifies head poses even with masks, addressing a critical limitation in face recognition.
• The integration of color space information, contour analysis, and deep learning offers a promising solution for robust head pose estimation.