Palmprint and face multi-modal biometric recognition based on SDA-GSVD and its kernelization.

Area of Science:

• Multimodal Biometrics
• Machine Learning
• Pattern Recognition

Background:

• Current multimodal feature extraction methods often overlook data distribution.
• Effective feature extraction is crucial for accurate biometric recognition.

Purpose of the Study:

• To propose a novel multimodal feature extraction and recognition approach based on Subclass Discriminant Analysis (SDA).
• To enhance biometric recognition by considering data distribution and treating different biometric modalities as subclasses of a single class.

Main Methods:

• Proposed Subclass Discriminant Analysis (SDA) where different biometric data (e.g., face, palmprint) from one person are treated as subclasses.
• Addressed singularity issues using PCA preprocessing and Generalized Singular Value Decomposition (GSVD).
• Developed nonlinear extensions: Kernel PCA-SDA (KPCA-SDA) and Kernel SDA with GSVD (KSDA-GSVD) for feature fusion.

Main Results:

• SDA-based approaches demonstrated superior performance compared to existing multimodal biometrics recognition methods.
• KSDA-GSVD achieved the highest recognition performance among the proposed methods.
• Experimental validation using palmprint and face data confirmed the effectiveness of the proposed techniques.

Conclusions:

• The proposed SDA framework effectively extracts discriminative features from multimodal biometric data.
• The nonlinear extensions, particularly KSDA-GSVD, offer significant improvements in multimodal biometric recognition accuracy.
• The method's ability to model data distribution as subclasses enhances its robustness and performance.