Offline Handwritten Signature Identification based on Hybrid Features and Proposed Deep Model

作     者：Hashim, Zainab Mohsin, Hanaa Alkhayyat, Ahmed 

作者机构：Department of Computer Sciences University of Technology Baghdad Iraq Department of Computer Technical Engineering College of Technical Engineering Islamic University Najaf Iraq 

出 版 物：《Iraqi Journal for Computer Science and Mathematics》 (Iraqi. J. Comput. Sci. Math.)

年 卷 期：2024年第5卷第1期

页      面：220-235页

核心收录：

主　　题：Textures 

摘      要：Handwritten signature identification is the process of determining an individual’s true identity by analyzing their signature. This is an important task in various applications such as financial transactions, legal document verification, and biometric systems. However, verifying handwritten signatures is challenging even in the age of digital transactions and remote document authentication. The inherent variety in people’s signatures, which may occur due to factors such as mood, exhaustion, or even the writing tool used, contributes to the problem. Furthermore, the proliferation of sophisticated forgery methods, such as freehand mimicking and sophisticated picture manipulation, necessitates the development of reliable and precise tools for identifying authentic signatures from fake ones. Various techniques have been developed for signature identification, including feature-based and machine learning-based methods. This paper proposes an authentic signature identification method based on integrating static (offline) signature data and a deep-based model, which fuses three types of signature features—Linear Discriminant Analysis as appearance-based features, Fast Fourier Transform as frequency-based features, and Grey-Level Co-occurrence Matrix as texture-based features. The fused features are then fed into the proposed deep-based model of 25 layers to identify each person. For experiments, we employed three datasets: our private collected dataset, called SigArab, and two public datasets called SigComp2011 and CEDAR. The proposed deep model achieved 99.23%, 100%, and 100% accuracy on the SigArab, CEDAR, and SigComp2011 datasets, respectively. In terms of precision, recall, and F-score, the findings revealed positive results for both datasets and exceeded 1.00, 0.487, and 0.655, respectively, on Sigcomp2011 dataset and 1.00, 0.507, and 0.672, respectively, on CEDAR dataset. © 2024 College of Education, Al-Iraqia University. All rights reserved.