Jointly Heterogeneous Palmprint Discriminant Feature Learning

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Heterogeneous palmprint recognition is crucial for advanced personal authentication.
• Existing methods often rely on hand-crafted features and lack cross-modal learning.
• Improving recognition performance across different palmprint types remains a significant challenge.

Purpose of the Study:

• To propose a novel simultaneous feature learning and encoding method for heterogeneous palmprint recognition.
• To develop a general model applicable to multiple heterogeneous palmprint recognition scenarios.
• To overcome limitations of traditional hand-crafted feature extraction.

Main Methods:

• Automatic learning of discriminant binary codes from direction convolution difference vectors.
• Joint feature learning across heterogeneous palmprint images to exploit modality-specific properties.
• Development of a general discriminative feature learning model for multi-modal recognition.

Main Results:

• The proposed method effectively learns discriminant features from heterogeneous palmprint data.
• Joint learning significantly improves the exploitation of unique properties from different modalities.
• Experimental validation on the PolyU multispectral palmprint database demonstrates superior performance.

Conclusions:

• The proposed simultaneous feature learning and encoding method offers a powerful approach for heterogeneous palmprint recognition.
• Jointly learning features from multiple modalities enhances recognition accuracy.
• The developed general model shows promise for diverse real-world biometric applications.