AuthSig: Safeguarding Scanned Signatures Against Unauthorized Reuse in Paperless Workflows

TL;DR

AuthSig tackles unauthorized reuse of scanned signatures in paperless workflows by binding authentication data to the signature through a diffusion-driven watermarking framework. It combines a keypoint-driven augmentation strategy with a content/style feature decoupling in a latent diffusion-based signature synthesizer, followed by a diffusion-based watermark embedding and extraction mechanism. The approach achieves high extraction accuracy (>98%) under diverse digital distortions and print-scan scenarios, with strong cross-media robustness and validated ablations confirming the importance of feature decoupling and VAE finetuning. This work enables a practical, intuitive signing-and-verification workflow that links authorization to specific documents, enhancing security for office and legal processes without sacrificing usability.

Abstract

With the deepening trend of paperless workflows, signatures as a means of identity authentication are gradually shifting from traditional ink-on-paper to electronic formats.Despite the availability of dynamic pressure-sensitive and PKI-based digital signatures, static scanned signatures remain prevalent in practice due to their convenience. However, these static images, having almost lost their authentication attributes, cannot be reliably verified and are vulnerable to malicious copying and reuse. To address these issues, we propose AuthSig, a novel static electronic signature framework based on generative models and watermark, which binds authentication information to the signature image. Leveraging the human visual system's insensitivity to subtle style variations, AuthSig finely modulates style embeddings during generation to implicitly encode watermark bits-enforcing a One Signature, One Use policy.To overcome the scarcity of handwritten signature data and the limitations of traditional augmentation methods, we introduce a keypoint-driven data augmentation strategy that effectively enhances style diversity to support robust watermark embedding. Experimental results show that AuthSig achieves over 98% extraction accuracy under both digital-domain distortions and signature-specific degradations, and remains effective even in print-scan scenarios.

Figures (7)

• Figure 1: System overview. Bob pre-registers n signature samples. For Alice's request, the system generates a metadata-bound signature valid only for this verification. Reuse with mismatched documents will fail authentication. The numbered arrows indicate the chronological sequence of operations.
• Figure 2: The training process of the proposed watermarking model. Embedding: Content feature $F_c$ is concatenated with noise latent $z_T$, while style feature $F_s$ is fused with watermark bits $w$ via a gated-residual block; this conditioned input guides the diffusion model to generate the watermarked signature. Extraction: Extractor $E_D$ employs STN for geometric alignment and a multi-layer MLP to recover $w$. A distortion layer $N$ enhances training robustness.
• Figure 3: Structural augmentation pipeline: the input signature is skeletonized, keypoints are extracted, and a Thin Plate Spline (TPS) transformation is applied for topologically consistent deformation.
• Figure 4: Schematic diagram of watermark embedding process.
• Figure 5: Visual examples of watermarked images $I_{w}$ and residuals of the proposed scheme and the compared methods. left: cover image $I_0$. Residuals are calculated as $(I_{w}-I_{o}) \times 2+128$.