ColorVein: Colorful Cancelable Vein Biometrics

Yifan Wang, Jie Gui, Xinli Shi, Linqing Gui, Yuan Yan Tang, James Tin-Yau Kwok

TL;DR

This paper introduces ColorVein, a novel method for creating cancelable vein biometrics by colorizing grayscale vein images. This approach enhances information density and allows for customizable, secure templates that can be revoked and reissued, addressing critical privacy concerns in biometric systems.

What Problem Does This Paper Address?

Vein recognition is a highly secure biometric identification method, but the risk of biometric data leakage poses a significant threat to user privacy. Existing cancelable biometric (CB) techniques often destroy original features or are not specifically designed for vein biometrics, leading to potential performance degradation. The paper tackles the challenge of creating robust, secure, and revocable vein biometric templates without compromising identification accuracy, especially in the face of potential data breaches or the need to update templates.

Methods

The core innovation of ColorVein lies in its use of image colorization to enhance vein biometrics. Instead of relying solely on grayscale vein patterns, ColorVein transforms them into dynamic, controllable color representations. This is achieved through an interactive colorization scheme where users or administrators can define a pseudo-random color space by specifying hint points (location, number, and color). This process injects richer information into the vein images, expanding the feature space.

Key methodological components include:

  • Vein Segmentation: A deep learning-based encoder-decoder network (ResU-Net) is used for robust vein segmentation, producing binarized vein patterns.
  • Vein Colorization: A user-interactive colorization network, inspired by deep learning models, takes the binarized vein image and user-defined hint points to generate a colored vein image. This network is trained on large-scale image datasets to learn general colorization capabilities.
  • Secure Center (SC) Loss Function: A novel objective function is proposed for training the feature extraction model. SC loss aims to maximize the distance between legitimate enrolled users and various types of impostors (including those from stolen tokens or cross-application scenarios), enhancing security and reliability.
  • Feature Extraction: A ResNet50 backbone is used for deep feature extraction, generating fixed-length feature vectors for matching.

Datasets

The paper utilizes four publicly available vein image datasets for evaluation:

  • HKPU-FV: Finger vein dataset.
  • PUT-FV: Palm vein dataset.
  • PUT-WV: Wrist vein dataset.
  • PUJ-DHV: Dorsal hand vein dataset.

The datasets were partitioned into enrollment (training/testing) and stolen scenario (training/testing) sets to simulate real-world security threats.

Key Insights and Ideas

  • Colorization as a Security Enhancement: The paper demonstrates that injecting color information into vein biometrics not only enriches feature representation but also serves as a powerful mechanism for cancelable template generation. This is a novel approach for vein biometrics.
  • Controllable Color Space for Revocability: The ability for users to define the color mapping through hint points is crucial for revocability. By simply changing the hint points, new, unrelated cancelable templates can be generated, allowing for template revocation without re-enrolling the biometric.
  • Balancing Hint Points: The number of hint points significantly impacts recognition performance and unlinkability. While more hint points generally improve unlinkability and reduce EER, there's a saturation point (around 10-25 hint points) beyond which performance gains diminish, and in stolen scenarios, too many hint points can even be detrimental.
  • Secure Center Loss for Robustness: The proposed SC loss function is key to enhancing the security of the system by creating a clear separation between genuine and impostor scores, even in challenging scenarios like stolen tokens.

Key Results

  • Recognition Performance: ColorVein achieved the best recognition performance (lowest EER) across all four datasets in both normal and stolen scenarios, significantly outperforming baseline methods and other representative cancelable biometric schemes like block remapping, mesh warping, and biohashing. For instance, in the HKPU-FV dataset, the EER for ColorVein in the normal scenario was 0.797%, compared to 2.250% for the baseline.
  • Robustness: ColorVein demonstrated stable performance on low-quality vein images, proving its robustness to image degradation.
  • Security Attributes:
    • Irreversibility: High privacy leakage rates (around 0.95-0.99) indicate that original biometric information cannot be reconstructed from ColorVein templates.
  • Brute Force Attack Defense: The system is computationally infeasible to attack via brute force due to the large feature space.
  • False Accept Attack Defense: Significant inference of feature bits (over 60%) is required for a false accept attack to succeed.
  • Unlinkability: Low global unlinkability scores (Dsys close to 0) indicate that templates generated with different tokens are effectively unlinked.
  • Revocability: High decidability index (d') values between genuine and pseudo-impostor distributions, and low values between impostor and pseudo-impostor distributions, confirm strong revocability.
  • Template Generation Time: ColorVein's template generation time (21.77 ms on average) is competitive, being slightly higher than some methods but offering superior security and performance.

Limitations and Future Work

  • Hint Point Optimization: While the paper explores the effect of hint point numbers, further research could focus on more automated or adaptive methods for hint point selection to optimize performance and unlinkability across different users and scenarios.
  • Computational Cost: Although competitive, the template generation time is slightly higher than some simpler methods. Future work could explore further optimizations in the colorization and feature extraction networks.
  • Dataset Diversity: While four datasets were used, expanding evaluation to more diverse vein datasets and imaging conditions could further validate the robustness of ColorVein.
  • User Experience: The interactive nature of colorization offers control, but streamlining this interaction for end-users could be an area for future development.

Summary updated

Abstract

Vein recognition technologies have become one of the primary solutions for high-security identification systems. However, the issue of biometric information leakage can still pose a serious threat to user privacy and anonymity. Currently, there is no cancelable biometric template generation scheme specifically designed for vein biometrics. Therefore, this paper proposes an innovative cancelable vein biometric generation scheme: ColorVein. Unlike previous cancelable template generation schemes, ColorVein does not destroy the original biometric features and introduces additional color information to grayscale vein images. This method significantly enhances the information density of vein images by transforming static grayscale information into dynamically controllable color representations through interactive colorization. ColorVein allows users/administrators to define a controllable pseudo-random color space for grayscale vein images by editing the position, number, and color of hint points, thereby generating protected cancelable templates. Additionally, we propose a new secure center loss to optimize the training process of the protected feature extraction model, effectively increasing the feature distance between enrolled users and any potential impostors. Finally, we evaluate ColorVein's performance on all types of vein biometrics, including recognition performance, unlinkability, irreversibility, and revocability, and conduct security and privacy analyses. ColorVein achieves competitive performance compared with state-of-the-art methods.
BibTeX
@article{wang2025colorvein,
  author = {Yifan Wang and Jie Gui and Xinli Shi and Linqing Gui and Yuan Yan Tang and James Tin-Yau Kwok},
  title = {{ColorVein: Colorful Cancelable Vein Biometrics}},
  journal = {IEEE Trans. Inf. Forensics Secur.},
  year = {2025},
}
Figure