An Improved Recognition Approach for Noisy Multispectral Palmprint by Robust L2 Sparse Representation with a...

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Multispectral palmprint recognition offers rich spatial and spectral data, surpassing single-spectral methods.
• Existing recognition technologies face challenges with noise and high-dimensional data.

Purpose of the Study:

• To develop an innovative and robust algorithm for multispectral palmprint recognition.
• To enhance recognition accuracy and resilience against noise contamination.

Main Methods:

• A robust L2 sparse representation (RL2SR) model incorporating a logistic function to evaluate residuals and suppress noise.
• A novel weighted sparse and collaborative concentration index (WSCCI) for adaptive fusion weight calculation.
• A tensor-based extreme learning machine (TELM) for direct high-dimensional data processing and spatial information preservation.

Main Results:

• The proposed RL2SR-TELM algorithm demonstrated superior performance on the PolyU benchmark multispectral palmprint database.
• The algorithm effectively handles both noise-free and various noise-contaminated multispectral palmprint images.
• RL2SR-TELM outperformed several state-of-the-art multispectral palmprint recognition algorithms.

Conclusions:

• The RL2SR-TELM algorithm presents a significant advancement in multispectral palmprint recognition.
• The adaptive fusion strategy and robust sparse representation contribute to improved accuracy and noise resistance.
• This method offers a promising solution for reliable biometric identification using multispectral palmprints.