3D Palmprint Identification Using Block-Wise Features and Collaborative Representation.

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• 3D palmprint recognition systems are gaining research interest due to unique advantages over 2D methods.
• Existing 3D palmprint matching is primarily for verification (one-to-one), lacking efficiency for identification (one-to-many).

Purpose of the Study:

• To address the limitations of existing methods for one-to-many 3D palmprint identification.
• To propose a novel collaborative representation (CR) based framework for efficient and accurate 3D palmprint identification.

Main Methods:

• Developed a CR-based framework using l1-norm or l2-norm regularizations for 3D palmprint identification.
• Introduced a block-wise statistics based feature extraction scheme using histograms of surface types.
• Evaluated the impact of different regularization terms and feature extraction methods.

Main Results:

• The CR-based framework with l2-norm regularization significantly improved recognition accuracy compared to other methods.
• The proposed feature extraction method yields highly discriminative and misalignment-robust feature vectors.
• The system demonstrated extremely low computational complexity, suitable for large-scale identification.

Conclusions:

• The proposed CR-based framework offers an efficient solution for one-to-many 3D palmprint identification.
• The block-wise statistics feature extraction enhances discriminative power and robustness.
• This approach is highly suitable for practical, large-scale biometric identification systems.