Abstract
An offline signature-based fuzzy vault (OSFV) is a bio-cryptographic implementation that uses handwritten signature images as biometrics instead of traditional passwords to secure private cryptographic keys. Having a reliable OSFV implementation is the first step towards automating financial and legal authentication processes, as it provides greater security of confidential documents by means of the embedded handwritten signatures. This paper reviews the first OSFV implementation and proposes new variants for enhanced accuracy and security. Specifically, a new method that adapts user key size is presented. Performance evaluations on the Brazilian PUCPR and GPDS signature databases indicate that the key-size adaptation method achieves a superior compromise between security and accuracy, increasing average system entropy from 45-bits to 51-bits while decreasing the average error rate (AER) by approximately 21%.
Introduction
In digital security, a fundamental tension exists between the exactness required by passwords and the inherent variability of biometrics. Biometric cryptosystems aim to bridge this gap. The "Fuzzy Vault" (FV) is a prominent scheme for this purpose, but applying it to offline signatures presents unique challenges. Unlike physiological biometrics (e.g., fingerprints) or dynamic online signatures, offline signatures are static images with high intrinsic variability. This "noise" makes it difficult to reconstruct a stable cryptographic key without compromising security or user experience.
System Architecture
The proposed OSFV pipeline consists of two primary phases: Enrollment and Authentication. During enrollment, a user's signature samples are processed to select a reliable representation and encode a secret key into a fuzzy vault. During authentication, a query signature is used to attempt to unlock the vault and retrieve the key.
Methodology
Dissimilarity Representation and Adaptive Matching
To handle signature variability, the system employs a dissimilarity representation. Instead of raw features, it calculates the distance between a query and a stored template, mapping signatures into a space where genuine samples cluster. Furthermore, the system utilizes a variability matrix ($\Delta$) to create custom matching windows for different features, tailoring the tolerance to the specific consistency of a user's signature.
The Fuzzy Vault Core
The secret key is hidden within a polynomial $p(f)$. Genuine points derived from the signature are used to evaluate this polynomial. To obscure these points, thousands of "chaff" (noise) points are added. To unlock the vault, a user must provide enough genuine points from a query signature to reconstruct the polynomial via Reed-Solomon decoding.
Adaptive Key Size
The core contribution of this work is the adaptive key size approach. Rather than using a fixed-degree polynomial, the system analyzes signature stability during enrollment:
- Stable signers are assigned a higher-degree polynomial, resulting in a longer, more secure key.
- Variable signers receive a lower-degree polynomial, ensuring the key remains reconstructible despite higher noise.
Experimental Evaluation and Results
The system was evaluated using the PUCPR (Brazilian) database (300 users) and the GPDS-300 public benchmark.
Key results include:
- Error Reduction: On the PUCPR database, the Average Error Rate (AER) dropped from 10.08% to 7.94%, a relative improvement of 21%.
- Enhanced Security: Average system entropy increased from 45 bits to 51 bits.
- State-of-the-Art Performance: The system achieved an AER of 16.75% on the GPDS database, comparable to complex traditional signature verification systems.
Conclusion
The adaptive key size logic successfully turns biometric variability into a parameter for security optimization. While the system is robust against random attacks, future work will focus on improving resistance to skilled simulated forgeries and exploring the application of this adaptive logic to other fuzzy biometrics such as gait or voice recognition.
Summary updated