A Fechner multiscale local descriptor for face recognition

Area of Science:

• Computer Vision
• Biometric Recognition
• Machine Learning

Background:

• Traditional feature extraction methods often struggle with variations in illumination, pose, and expression.
• Human visual perception offers a powerful model for robust feature representation.
• Fechner's law describes the relationship between physical stimuli and perceived intensity.

Purpose of the Study:

• To introduce a new feature descriptor, the Fechner multiscale local descriptor (FMLD), inspired by Fechner's law.
• To enhance face recognition accuracy by simulating human pattern perception.
• To improve the performance of convolutional neural networks (CNNs) in face recognition tasks.

Main Methods:

• FMLD employs multiscale local domains to capture structural facial features, simulating human perception of intensity differences.
• It extracts magnitude and direction features using binary patterns, maintaining a close relationship between them.
• Feature maps are fused into an overall histogram for comprehensive representation.

Main Results:

• FMLD demonstrates robust performance in face recognition, effectively handling variations in illumination, pose, expression, and occlusion.
• The descriptor significantly enhances the performance of CNNs when integrated.
• The combined FMLD and CNN approach outperforms existing advanced descriptors.

Conclusions:

• FMLD offers a novel and effective approach to feature extraction for face recognition by leveraging principles of human perception.
• The descriptor's ability to capture intricate facial details and its compatibility with CNNs make it a valuable tool for biometric systems.
• FMLD represents a significant advancement in addressing real-world challenges in face recognition.