TL;DR
This paper introduces a novel geometric transformation to secure fingerprint templates by converting minutiae point configurations into a fixed-length code. This approach aims to protect irreplaceable biometric data from compromise while enabling efficient matching and alignment, offering a practical solution to template security concerns.
What Problem Does This Paper Address?
The increasing reliance on biometrics, particularly fingerprints, for authentication presents a critical security challenge: biometric data is irreplaceable. If a fingerprint template stored in a database or on a device is compromised, it cannot be revoked or changed like a password. This makes biometric templates highly attractive targets for attackers. Traditional security methods like cryptographic hashing are unsuitable for biometric data due to its inherent variability and the need for error tolerance during matching. The paper specifically addresses the problem of securing minutiae-based fingerprint templates, which are representations of key fingerprint features like ridge endings and bifurcations.
Methods
The proposed scheme transforms a set of minutiae points from a fingerprint into a fixed-length "fingerprint code." The core steps involve:
- Extraction: Identifying minutiae points from a fingerprint image.
- Mapping:
- Calculating the centroid of the minutiae points.
- Drawing a circle centered at this centroid.
- For pairs of minutiae points further apart than a predefined threshold, lines are drawn through them.
- The intersection points of these lines with the circle are determined.
- Quantization: The circle is divided into a specified number of bins (determined by the parameter $\Delta$), and the intersection points are assigned to these bins based on their angular position.
- Code Generation: The number of intersection points falling into each bin is concatenated to form the fingerprint code. This results in a fixed-length vector, regardless of the original number of minutiae points.
This transformation is designed to be a robust, one-way function, meaning it's difficult to reconstruct the original minutiae points from the generated code. Matching is then performed by comparing these fingerprint codes using a mean absolute error metric.
Datasets
The paper does not use publicly available datasets. Instead, it generates synthetic fingerprint data for its experiments. Specifically, it creates 800 synthetic fingerprints, each with 20 randomly generated minutiae points. To evaluate the robustness of the scheme, three distinct test datasets are generated:
- Noisy Versions (Pixel Displacement): 20 noisy versions of each original fingerprint are created by displacing minutiae points by up to 5 pixels in x and/or y directions.
- Noisy Versions (Minutiae Addition/Deletion): 20 noisy versions are created by adding or removing up to 2 minutiae points from the original fingerprint.
- Rotated Versions: Fingerprints are randomly rotated by up to 30 degrees (clockwise and counter-clockwise) to test alignment robustness.
Key Insights and Ideas
- Geometric Transformation for Security: The core idea of using geometric relationships between minutiae points and mapping them onto a circle is a novel approach to template security. It moves beyond simple hashing or encryption by leveraging the spatial arrangement of features.
- Fixed-Length Code for Efficiency: Generating a fixed-length code simplifies storage and matching, making the system more efficient.
- Robustness to Alignment: The cyclic nature of the fingerprint code means that misalignments in query fingerprints can manifest as cyclic shifts in the code, which can be detected and potentially corrected.
- One-Way Property: The combination of geometric transformation, intersection point calculation, and binning is intended to make it computationally infeasible to reverse-engineer the original minutiae locations from the generated code.
Key Results
The paper presents experimental results using synthetic data, evaluating the scheme's performance in terms of Equal Error Rate (EER) across different parameter settings ($\Delta$ for quantization bin size and $T$ for distance threshold).
- Parameter Sensitivity: The performance (EER) is shown to be sensitive to the choice of $\Delta$ and $T$. Generally, very small or very large values for these parameters lead to poorer performance.
- Optimal Parameters: For the first test dataset (pixel displacement), an optimal $\Delta$ of 8 degrees (resulting in 45 bins) and a $T$ of 80 pixels yielded the lowest EER.
- Impact of Noise: The scheme is more robust to small pixel displacements than to minutiae addition/deletion. The EER values are significantly higher for the dataset involving minutiae addition/deletion.
- Performance: For the first dataset, the best EER achieved was around $0.67 \times 10^{-4}$. For the second dataset (minutiae addition/deletion), the best EER was around $8.08 \times 10^{-3}$ with $\Delta=18$ and $T=120$.
Limitations and Future Work
- Synthetic Data: The experiments were conducted on synthetically generated minutiae points. The authors acknowledge that real-world fingerprint data, with its inherent complexities and variations, needs to be used for more realistic performance evaluation.
- Brute-Force Attack Security: While the scheme aims to make brute-force attacks difficult, the authors note that a more rigorous investigation into security and entropy loss under various attack scenarios is required.
- Alternative Geometries: The current scheme uses a circle. Future work includes exploring other geometric shapes (e.g., ellipses) and placement strategies.
- Real-World Data and Analysis: Extending experiments to real fingerprint data, analyzing alternative geometric shapes, and conducting more in-depth security analysis against various attack scenarios are identified as future research directions.
Summary updated