Triple-Type Feature Extraction for Palmprint Recognition

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Palmprint recognition is a user-friendly biometric technology due to its non-invasive and hygienic properties.
• Current deep-learning methods for palmprint recognition require extensive labeled data, which is often difficult to obtain.
• Limited availability of palmprint samples hinders the effectiveness of existing deep-learning-based recognition systems.

Purpose of the Study:

• To propose a novel heuristic palmprint recognition method that does not require training samples.
• To address the challenge of limited data availability in palmprint recognition systems.
• To develop an effective palmprint recognition technique that utilizes inherent palmprint features.

Main Methods:

• Extraction of three inherent palmprint features: texture, gradient, and direction.
• Encoding these features into triple-type feature codes.
• Representation using block-wise histograms of feature codes to form triple feature descriptors.
• Weighted matching-score level fusion for similarity calculation and recognition.

Main Results:

• The proposed heuristic method demonstrates promising effectiveness in palmprint recognition.
• The method achieves good performance without the need for any training samples.
• Experimental results on three widely used palmprint databases validate the method's efficacy.

Conclusions:

• The developed heuristic palmprint recognition method is effective, particularly in scenarios with limited data.
• The triple-type feature extraction and fusion strategy provide a robust approach to palmprint identification.
• This method offers a viable alternative to data-intensive deep-learning techniques for palmprint recognition.