Privacy-Preserving Biometric Verification with Handwritten Random Digit String

TL;DR

The study tackles privacy concerns in online handwriting verification by introducing Random Digit Strings (RDS) as a privacy-preserving, text-independent biometric. It proposes PAVENet with a Discriminative Pattern Mining module to robustly capture personal writing patterns under highly variable content, and validates this approach on the new HRDS4BV TI-RDS dataset. The work shows substantial gains over prior TD/TI methods, highlights a surprising finding that skilled forgery can be easier to verify than random forgery in RDS, and discusses implications for privacy, security, and future research. Collectively, the results indicate that privacy-preserving handwriting verification using RDS is feasible and holds promise for widespread, privacy-conscious biometric applications.

Abstract

Handwriting verification has stood as a steadfast identity authentication method for decades. However, this technique risks potential privacy breaches due to the inclusion of personal information in handwritten biometrics such as signatures. To address this concern, we propose using the Random Digit String (RDS) for privacy-preserving handwriting verification. This approach allows users to authenticate themselves by writing an arbitrary digit sequence, effectively ensuring privacy protection. To evaluate the effectiveness of RDS, we construct a new HRDS4BV dataset composed of online naturally handwritten RDS. Unlike conventional handwriting, RDS encompasses unconstrained and variable content, posing significant challenges for modeling consistent personal writing style. To surmount this, we propose the Pattern Attentive VErification Network (PAVENet), along with a Discriminative Pattern Mining (DPM) module. DPM adaptively enhances the recognition of consistent and discriminative writing patterns, thus refining handwriting style representation. Through comprehensive evaluations, we scrutinize the applicability of online RDS verification and showcase a pronounced outperformance of our model over existing methods. Furthermore, we discover a noteworthy forgery phenomenon that deviates from prior findings and discuss its positive impact in countering malicious impostor attacks. Substantially, our work underscores the feasibility of privacy-preserving biometric verification and propels the prospects of its broader acceptance and application.

Figures (11)

• Figure 1: Visualization of our proposed online handwritten RDS and comparisons with signatures and TDS. The samples of Chinese signature and TDS are obtained from the MSDS dataset zhang2022msds, which are desensitized through mosaic to protect individual privacy.
• Figure 2: Warping paths of DTW between different query and template pairs. The left results are with a TDS pair of fixed content, whereas the right ones are with an RDS pair of different content. Smoother curve and smaller warping distance are better.
• Figure 3: Illustration of the construction details of the HRDS4BV dataset, including data collection devices, the graphical user interface of the data acquisition app, and the acquisition process.
• Figure 4: Data visualization. (a) Visualization of data diversity. The samples are randomly selected from distinct writers, showcasing obvious inter-writer variations in writing styles. (b) Visualization of data representativeness. We exhibit all genuine samples and skilled forgeries from a random writer. The contents and lengths of genuine samples vary since they are randomly assigned. Each skilled forgery holds a one-to-one correspondence to a specific genuine sample. The data were collected in two sessions separated by at least three weeks, with 10 genuine and 10 skillfully forged samples per session.
• Figure 5: Overall framework of the proposed PAVENet. Top: model training. The training data includes genuine RDS, the corresponding skilled forgeries per writer, and random forgeries. The model is trained to distinguish genuine and forged RDS through feature space optimization. Middle: a simplified model architecture of the PAVENet. The PAVENet is dedicatedly designed to capture discriminative handwriting patterns, bolstering the learning of unique writing styles. Bottom: model inference. The model compares the query RDS against the template to determine whether to accept or reject the query.