3D texture-based face recognition system using fine-tuned deep residual networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Rapid advancements in 3D photography have generated vast amounts of 3D image data.
• 3D face recognition is a key research area, but traditional methods struggle with large datasets, leading to decreased accuracy and inefficiency.
• Deep learning offers a promising approach to overcome these limitations in 3D face recognition.

Purpose of the Study:

• To propose an end-to-end face recognition system utilizing 3D face texture for enhanced accuracy.
• To investigate the effectiveness of combining geometric invariants, histogram of oriented gradients, and fine-tuned residual neural networks (ResNet) for 3D face recognition.
• To demonstrate the efficiency and improved performance of the proposed deep learning framework compared to traditional methods.

Main Methods:

• Developed an end-to-end 3D face recognition system integrating 3D face texture, geometric invariants, and histogram of oriented gradients.
• Employed fine-tuned residual neural networks (ResNet) for deep learning-based classification of 3D face data.
• Evaluated the system's performance using the FRGC-v2 dataset.

Main Results:

• The proposed system achieved a Top-1 accuracy of 98.26% and a Top-2 accuracy of 99.40% on the FRGC-v2 dataset.
• Increased layers in the fine-tuned ResNet deep neural network led to improved recognition accuracy.
• The deep learning framework required fewer iterations compared to traditional 3D face recognition methods.

Conclusions:

• The proposed deep learning framework significantly enhances 3D face recognition accuracy, especially with large datasets.
• The integration of 3D face texture, geometric invariants, and fine-tuned ResNet offers a robust solution for realistic 3D face recognition scenarios.
• This approach provides a more efficient and accurate method for classifying large volumes of 3D face data.