Bimodal Biometric Verification Using the Fusion of Palmprint and Infrared Palm-Dorsum Vein Images

Chih-Lung Lin, Shih-Hung Wang, Hsu-Yung Cheng, Kuo-Chin Fan, Wei-Lieh Hsu, Chin-Rong Lai2015 · DOI

TL;DR

This paper presents a novel bimodal biometric verification system that fuses palmprint and infrared palm-dorsum vein images. By combining these two modalities, the system achieves higher accuracy and robustness than unimodal systems, offering a promising approach for secure identity verification.

What Problem Does This Paper Address?

Unimodal biometric systems, while convenient, often face challenges such as limited information, susceptibility to spoofing, and insufficient accuracy. To overcome these limitations, bimodal biometric systems that combine multiple biometric traits have emerged. This paper specifically addresses the need for a more reliable and user-friendly biometric system by fusing palmprint and palm-dorsum vein patterns. The challenge lies in effectively integrating these two distinct modalities, which have different image characteristics and acquisition requirements, to achieve superior performance.

Methods

The proposed method involves several key steps:

  • Image Acquisition: Palmprint images are captured using a low-resolution digital scanner, and palm-dorsum vein images are captured using an infrared (IR) camera. This peg-free approach enhances user-friendliness.
  • Preprocessing: This stage includes palm region segmentation, identification of finger-web locations for alignment, and cropping of a standardized Region of Interest (ROI).
  • Image Fusion: The core of the method involves fusing the palmprint and vein pattern ROIs using the Discrete Wavelet Transform (DWT) and Inverse Discrete Wavelet Transform (IDWT). A novel "Avg-Max" hybrid fusion rule is introduced, which averages approximation coefficients and takes the maximum of detail coefficients. This rule is designed to leverage the complementary information from both modalities.
  • Feature Extraction: Line-like features (LLFs) are extracted from the fused image using Sobel operators to enhance edges. An iterative thresholding technique is employed to accurately segment these LLFs.
  • Multiresolution Analysis: A multiresolution filter (feature-pass filter) is used to decompose the LLFs into different resolutions, creating feature vectors that are robust to positional offsets.
  • Verification: A Support Vector Machine (SVM) with a radial basis function kernel is used for the final verification of identity.

Datasets

The study utilized a dataset collected from 100 volunteers, comprising 3000 palmprint images and 3000 palm-dorsum vein images. This dataset was collected using a peg-free scenario to enhance user-friendliness.

Key Insights and Ideas

  • Complementary Modalities: Palmprints and palm-dorsum veins offer distinct physiological characteristics, and their fusion effectively compensates for the weaknesses of individual modalities.
  • IR Imaging for Veins: The use of IR imaging for vein patterns leverages the temperature difference between veins and skin, providing a robust and antispoofing mechanism.
  • Hybrid Fusion Rule: The "Avg-Max" fusion rule is a novel approach that effectively combines the approximation and detail coefficients from DWT, tailored to the different frequency characteristics of palmprints and vein patterns.
  • Peg-Free Acquisition: Eliminating the need for pegs significantly improves the user-friendliness and practicality of the biometric system.
  • Low-Resolution Advantage: Using low-resolution images leads to a smaller database size, which is beneficial for storage and processing efficiency.

Key Results

The proposed bimodal biometric verification system demonstrated promising results:

  • False Rejection Rate (FRR): 1.20%
  • False Acceptance Rate (FAR): 1.56%

These results indicate a high level of accuracy and reliability, outperforming many unimodal biometric systems. The paper also shows comparative results against other methods in Table 1, demonstrating its competitive performance.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • IR Camera Resolution: The low resolution and sensibility of the IR camera used in the study limit the accuracy of biometric verification. A higher-performance IR camera is recommended for capturing higher-quality and more discriminative images.
  • Feature Extraction: Exploring other effective feature extraction methods could potentially yield more information from palm and palm-dorsum images.
  • Multimodal Fusion: Incorporating additional biometric features, such as palm geometry, fingerprints, or palm creases, could further enhance verification accuracy.
  • Feature Variation: Biometric features can change with age. Future work could focus on developing methods to predict and account for these feature variations to maintain long-term accuracy.

Summary updated

BibTeX
@misc{lin2015bimodal,
  author = {Chih-Lung Lin and Shih-Hung Wang and Hsu-Yung Cheng and Kuo-Chin Fan and Wei-Lieh Hsu and Chin-Rong Lai},
  title = {{Bimodal Biometric Verification Using the Fusion of Palmprint and Infrared Palm-Dorsum Vein Images}},
  year = {2015},
  doi = {10.3390/s151229856},
}
Figure