An Effective Palmprint Recognition Approach for Visible and Multispectral Sensor Images

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Histogram of Oriented Gradients (HOG) is effective for palmprint feature extraction but struggles with rotational variations.
• Existing methods lack robustness in recognizing palmprints under different rotation angles.

Purpose of the Study:

• To develop a robust palmprint recognition approach by combining HOG with Steerable Gaussian Filters (SGF).
• To improve feature extraction accuracy and recognition performance, especially under rotational variations.

Main Methods:

• A hybrid HOG-SGF feature extraction method was developed.
• Palmprint images were segmented using David Zhang's method.
• Feature dimensionality was reduced using an optimized Auto-Encoder (AE).
• Classification was performed using a Regularized Extreme Learning Machine (RELM).

Main Results:

• The proposed HOG-SGF method demonstrated superior performance compared to existing state-of-the-art approaches.
• Effective palmprint recognition was achieved even with a small number of training samples.
• Robustness against rotation was significantly improved.

Conclusions:

• The hybrid HOG-SGF method offers a robust and effective solution for palmprint recognition.
• This approach advances the field of biometrics by addressing key limitations in feature extraction.
• The method shows promise for real-world applications requiring high accuracy and robustness.