Offline Handwritten Signature Verification Using a Stream-Based Approach

Kecia G. de Moura,Rafael M. O. Cruz,Robert Sabourin

2024-11-11

Abstract:Handwritten Signature Verification (HSV) systems distinguish between genuine and forged signatures. Traditional HSV development involves a static batch configuration, constraining the system's ability to model signatures to the limited data available. Signatures exhibit high intra-class variability and are sensitive to various factors, including time and external influences, imparting them a dynamic nature. This paper investigates the signature learning process within a data stream context. We propose a novel HSV approach with an adaptive system that receives an infinite sequence of signatures and is updated over time. Experiments were carried out on GPDS Synthetic, CEDAR, and MCYT datasets. Results demonstrate the superior performance of the proposed method compared to standard approaches that use a Support Vector Machine as a classifier. Implementation of the method is available at <a class="link-external link-https" href="https://github.com/kdMoura/stream_hsv" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition,Artificial Intelligence,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to adapt to the inherent variability of handwritten signatures and their characteristics evolving over time in a handwritten signature verification system, while maintaining high verification performance and alleviating the problem of limited data. Specifically, traditional off - line handwritten signature verification (HSV) systems are usually developed based on static batch configurations, which limit the system's ability to model signature changes. Signatures have a high degree of intra - class variability and are easily affected by time and other external factors, so a method that can dynamically adapt to these changes is required.

To solve this problem, the author proposes a new data - stream - based handwritten signature verification framework (StreamHSV), which treats signatures as infinite data streams and continuously updates and improves the system over time. The following are the main contributions of this framework:

1. **StreamHSV**: A new HSV framework is proposed, which can adapt over time and achieve continuous learning and improvement by treating signatures as infinite data streams.
2. **Stream Dichotomy Transformation**: A stream dichotomy transformation process is introduced to promote adaptive learning from the incoming signature stream and solve the common data imbalance problem in the stream scenario.
3. **Signature Stream Generation Method**: In order to evaluate using standard batch processing configurations, a method for generating signature streams based on existing HSV evaluation protocols is proposed.

Through these improvements, this research aims to improve the accuracy and reliability of handwritten signature verification systems, especially in the case of limited data.

### Formula Summary

- **Feature Extraction Model**:
\[
x_C=\phi(S_C)
\]
where \(S_C\) is the signature image of the claimed user and \(x_C\) is the corresponding feature vector.

- **Stream Dichotomy Transformation**:
\[
\dot{x}_{RC} = DT(x_R, x_C)=\{|f_{Rk}-f_{Ck}|\}_{k = 1}^K
\]
where \(x_R\) and \(x_C\) are the feature vectors of the reference signature and the claimed signature respectively, and \(K\) is the number of features.

- **Loss Function (for SGD Classifier)**:
\[
\min_{w,b}\frac{C}{2}\|w\|^2+\frac{1}{N}\sum_{i = 1}^{N}\max(0,1-y_i(w\cdot x_i + b))
\]
where \(C\) is the regularization parameter, \(N\) is the number of samples, \(y_i\) is the label, \(w\) is the weight vector, and \(b\) is the bias term.

Through these methods, the paper provides a more flexible and adaptable handwritten signature verification solution.