Intelligent multimodal 3D biometric recognition using PointNet + + for robust face-ear authentication.

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• 2D biometric systems face reliability issues due to illumination, expressions, and occlusion.
• 3D biometrics offer more robust structural information and environmental resilience.
• Multimodal biometrics combining 3D face and 3D ear data can improve recognition accuracy.

Purpose of the Study:

• To develop and evaluate a multimodal biometric recognition system using 3D face and 3D ear data.
• To leverage the PointNet++ model for feature extraction from 3D point clouds.
• To enhance the reliability and accuracy of biometric identity recognition.

Main Methods:

• Applied pre-processing techniques including cropping, normalization, hole filling, and spike removal to 3D biometric data.
• Utilized the PointNet++ model, a Convolutional Neural Network (CNN) variant, for direct processing of 3D point clouds.
• Tested the system using the 3D Face Recognition Grand Challenge (FRGC) database and the University of Notre Dame (UND) Collection G database.

Main Results:

• PointNet++ achieved 99% accuracy for 3D face recognition.
• PointNet++ achieved 98% accuracy for 3D ear recognition.
• The model demonstrated high accuracy by learning multi-scale features for both local and global information.

Conclusions:

• The proposed multimodal biometric system effectively integrates 3D face and 3D ear data for reliable identity recognition.
• The PointNet++ model's 3D point cloud optimization and resilient architecture are key to achieving high accuracy.
• This approach offers a significant improvement over traditional 2D biometric systems, particularly in challenging real-world conditions.