Tensorized Multi-View Low-Rank Approximation Based Robust Hand-Print Recognition.

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Hand-print recognition (palmprint, finger-knuckle-print, hand-vein) offers user convenience and hygiene advantages.
• Existing multi-view methods struggle with noise, rotation, and shadow, and often fail to capture high-order correlations between views.

Purpose of the Study:

• To develop a robust hand-print recognition method that addresses limitations of existing multi-view approaches.
• To enhance feature representation by effectively utilizing multi-view information and modeling cross-view correlations.

Main Methods:

• Proposed a novel tensorized multi-view low-rank approximation based robust hand-print recognition method (TMLA_RHR).
• Formulated the method using aligned structure regression loss and tensorized low-rank approximation within a joint learning model.
• Treated low-rank representation matrices of different views as a tensor, regularized by a low-rank constraint to model cross-view information and reduce redundancy.

Main Results:

• The TMLA_RHR method demonstrated superior performance in hand-print recognition tasks.
• Experimental results on eight real-world databases confirmed the method's effectiveness.
• The approach successfully handled interference factors like noise and rotation.

Conclusions:

• The proposed TMLA_RHR method offers a significant advancement in robust hand-print recognition.
• Tensorized low-rank approximation effectively models multi-view correlations, leading to compact and discriminative feature representations.
• The method shows promise for practical applications requiring reliable biometric identification.