Enhanced Multitask Learning for Hash Code Generation of Palmprint Biometrics.

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Palmprint recognition is crucial for secure identification.
• Existing methods often struggle with balancing accuracy and template storage efficiency.
• Multitask learning offers potential for improved performance by joint optimization.

Purpose of the Study:

• To propose a novel multitask learning framework for palmprint biometrics.
• To jointly optimize classification (identity, gender, chirality) and hashing tasks.
• To enhance palmprint template storage and matching efficiency.

Main Methods:

• Developed a multitask learning framework with joint classification and hashing branches.
• Integrated an attention mechanism module for channel weighting.
• Incorporated a customized gate control module for expert knowledge integration.
• Implemented an automatic weight adjustment module for task optimization.

Main Results:

• The framework achieved superior performance compared to isolated tasks.
• Demonstrated promising accuracies across various classification tasks.
• Significantly improved palmprint authentication accuracy.
• Validated the efficacy of the integrated attention, gate control, and weight adjustment modules.

Conclusions:

• The proposed multitask learning framework effectively enhances palmprint biometric system performance.
• Joint optimization of classification and hashing, along with novel modules, leads to improved accuracy and efficiency.
• The framework shows significant potential for real-world biometric applications.