Online Signature Watermarking in the Transform Domain

Artificial Intelligence Computer Science Information Science
online signature transform methods watermarking multi-bit watermarking digital signature biometrics security

Academic Background

With the rapid growth of digital content, the importance of digital signatures in identity verification and content authentication has become increasingly prominent. However, the security and integrity of digital signatures face significant challenges. To protect the authenticity of signatures and prevent tampering, digital watermarking technology has emerged. Digital watermarking embeds invisible yet identifiable information into digital content, effectively verifying the source and integrity of data. In recent years, transform-domain watermarking techniques, such as the Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT), have gained attention for their ability to balance robustness, imperceptibility, and recognition accuracy.

This paper explores multi-bit watermarking techniques for online handwritten signatures, particularly focusing on transform-domain methods based on DCT and DWT. The study investigates how to balance signal distortion, watermark extraction accuracy, and biometric recognition rates during watermark embedding, providing solutions for the security and robustness of online signature biometrics.

Source of the Paper

This paper was authored by Marcos Faundez-Zanuy from the Technology Department of Tecnocampus, affiliated with Universitat Pompeu Fabra in Barcelona, Spain. The paper was published in 2025 in the journal Cognitive Computation, with the DOI 10.1007/s12559-025-10436-y.

Research Process

1. Database and Experimental Subjects

The study utilized the MCYT signature database, which contains online handwritten signature data from 330 participants. Each participant provided 25 genuine signatures and 25 forged signatures. The signature data was captured using a Wacom Intuos A6 USB tablet, recording pen position (x, y coordinates), pressure (p), and pen azimuth and altitude angles.

2. Watermark Embedding and Extraction

DCT Watermarking Technique

  • Watermark Embedding: The signature signal is first transformed into the frequency domain using the two-dimensional Discrete Cosine Transform (DCT2). A randomly generated binary watermark is then embedded into the DCT coefficients, with the watermark strength controlled by the parameter α. Finally, the signal is transformed back to the spatial domain using the inverse DCT to obtain the watermarked signature signal.
  • Watermark Extraction: The watermark is extracted by comparing the DCT coefficients of the original and watermarked signals using the formula ( wr = \frac{dct{watermarked} - dct_{original}}{\alpha} ).

DWT Watermarking Technique

  • Watermark Embedding: The signature signal is decomposed using the Haar wavelet to obtain approximation and detail coefficients. The watermark is embedded into the approximation coefficients, and the signal is reconstructed using the inverse wavelet transform.
  • Watermark Extraction: The watermarked signal is decomposed using the wavelet transform, and the watermark is extracted by comparing the approximation coefficients of the original and watermarked signals.

3. Parameter Optimization and Error Analysis

The study experimentally determined the minimum α values for DCT and DWT watermarking techniques as 2.56 and 0.71, respectively, to ensure error-free watermark extraction. Additionally, the study analyzed the impact of different α values on signal distortion and watermark extraction accuracy.

4. Multi-bit Watermark Embedding

The study further explored multi-bit watermark embedding techniques, where multiple bits of watermark information are embedded into each sample. Experiments validated the impact of multi-bit embedding on signal distortion and biometric recognition rates, and an optimal embedding strategy was proposed.

Main Results

  1. Watermark Extraction Accuracy: Experiments showed that DCT and DWT techniques perform excellently in watermark extraction, particularly when the α value exceeds a certain threshold, significantly reducing the error rate.
  2. Signal Distortion Analysis: As the α value increases, watermark robustness improves, but it also leads to increased distortion in the signature signal. DWT technology performs slightly better than DCT in terms of signal distortion.
  3. Biometric Recognition Rate: Within a moderate range of α values, watermark embedding has minimal impact on signature recognition accuracy and verification performance, demonstrating the feasibility of transform-domain watermarking in biometric recognition.
  4. Multi-bit Embedding Performance: Multi-bit embedding significantly increases watermark capacity but also increases signal distortion. The study determined the optimal number of embedding bits for each technique to balance watermark capacity and signal quality.

Conclusions and Significance

This paper systematically investigates the application of transform-domain watermarking techniques based on DCT and DWT in online handwritten signatures. The results demonstrate that transform-domain watermarking effectively balances watermark robustness and imperceptibility, providing reliable solutions for the security and integrity of online signature biometrics. Additionally, the exploration of multi-bit watermarking lays the foundation for future high-capacity watermarking applications.

Research Highlights

  1. Innovative Methods: This paper is the first to apply multi-bit watermarking techniques to online handwritten signatures, significantly improving watermark capacity and robustness.
  2. Comprehensive Experimental Analysis: Extensive experiments determined the optimal parameters for DCT and DWT techniques, providing theoretical support for practical applications.
  3. Interdisciplinary Application: The study combines digital watermarking and biometric recognition technologies, offering new research directions for the field of information security.

Other Valuable Information

The paper also explores the tolerance of watermarking techniques to noise, showing that watermarks can still be accurately extracted even when the signal is subjected to a certain level of interference. Additionally, the study provides detailed experimental data and algorithm implementation code, offering valuable reference resources for future research.

Through this study, the application prospects of transform-domain watermarking in the protection of online signature biometrics are further validated, providing important technical support for the design of future secure authentication systems.