Recognizing partially occluded, expression variant faces from single training image per person with SOM and soft...

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Classical face recognition often requires multiple training images per person.
• Real-world applications frequently face challenges like single-sample learning, occlusions, and expression variations.
• Existing methods struggle with these constraints, necessitating more robust approaches.

Purpose of the Study:

• To develop a robust frontal face recognition technique for single-sample learning scenarios.
• To address challenges posed by partial occlusions and varying expressions in test images.
• To improve the performance of face recognition systems in unconstrained environments.

Main Methods:

• Extended a local probabilistic approach using Self-Organizing Maps (SOMs) to learn individual-specific subspaces.
• Proposed two SOM training strategies: a single map for all samples and separate maps per class.
• Introduced a soft kappa nearest neighbor (soft kappa-NN) ensemble method for subject identification.

Main Results:

• The proposed SOM-based method demonstrated high robustness against partial occlusions.
• The technique effectively handled variations in facial expressions during recognition.
• Experimental results confirmed the superior performance compared to traditional methods in challenging conditions.

Conclusions:

• The SOM-based approach offers a significant advancement for single-sample face recognition.
• The method provides a robust solution for real-world face recognition applications with occlusions and expression variations.
• This work contributes to the development of more resilient and practical biometric systems.