Making the Most of Single Sensor Information: A Novel Fusion Approach for 3D Face Recognition Using Region Covariance...

Area of Science:

• Computer Vision
• Biometrics
• Pattern Recognition

Background:

• Commercial face recognition systems often use multi-sensor data for robustness.
• Single-sensor systems, particularly with 3D data, face challenges in recognition accuracy and reliability.

Purpose of the Study:

• To propose an automated 3D face recognition framework using a single 3D sensor.
• To enhance face recognition performance without relying on pre-annotated facial landmarks.

Main Methods:

• Utilizes region covariance matrixes and Gaussian mixture models (GMMs) for feature extraction and modeling.
• Employs the unscented transform for deriving low-dimensional feature vectors from local face image regions.
• Incorporates a support vector machine (SVM) classifier for final identity decision.

Main Results:

• Achieves competitive results on prominent 3D face databases (FRGC v2, CASIA, UMB-DB).
• Demonstrates robustness through inherent data fusion mechanisms within region covariance descriptors.
• Shows efficacy in exploring facial images at multiple locality levels.

Conclusions:

• The proposed automated framework effectively performs 3D face recognition using a single sensor.
• The method integrates region covariance, GMMs, and SVMs to achieve robust and accurate recognition.
• Normalization techniques further improve the performance of the 3D face recognition system.