International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024
p-ISSN: 2395-0072
www.irjet.net
Signature Forgery Detection Using Deep Learning P. BHUVANESWARI1, K. MELVIN CHRISTOPHER2, V. RISHI MAHESH RAJ3, M. AJITHKUMAR4 1234 Dept. of Computer Science and Engineering, Government College of Engineering Srirangam, Tamilnadu, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – Digital signatures are widely adopted by
respectively [5]. This study utilizes CNNs to enhance the accuracy and reliability of the proposed signature recognition system. The primary objective is to develop a deep learning-based system capable of not only identifying genuine signatures but also detecting forgeries, thereby reducing the need for manual intervention. This approach aims to save time and costs associated with traditional methods. To accomplish these objectives, the study focuses on assembling and preprocessing a comprehensive dataset of signatures [1]. Preprocessing steps include noise removal to facilitate the implementation of a deep learning architecture for signature recognition. Evaluation metrics such as accuracy, precision, recall, and F1 score are utilized, and the system's performance is benchmarked against other state-of-the-art methods. In summary, this study provides valuable insights into model performance, dataset requirements, and potential areas for improvement in the field of signature recognition. The findings underscore the significance of training on diverse datasets and emphasize the capabilities of deep learning approaches.
organizations, both public and private, in recent times due to their legal validity and ease of handling and storage. They find extensive usage in e-commerce websites for customer authentication during deliveries, bank procedures, government organizations, and various other businesses. Governments also utilize digital signatures for contract signing and document verification. However, with advancements in Information Technology (IT), there are both advantages and disadvantages. While digital signatures offer convenience, security, and cost savings, they also pose risks, such as potential forgery or manipulation. To address the risk of signature forgery, researchers are exploring deep learning algorithms like VGG16. These algorithms analyze signature data to differentiate between genuine and fake signatures by learning patterns and features from a dataset. By training and testing these algorithms on diverse signature samples, researchers aim to develop robust systems for detecting and mitigating signature forgery attempts. In summary, digital signatures play a vital role in modern organizational operations, offering benefits like legal validity, convenience, and enhanced security. However, addressing potential risks, such as forgery, requires ongoing research and technological advancements, including the application of deep learning algorithms like VGG16.
1.1 RELATED WORK A. In the field of handwritten signature identification, researchers frequently employed the ResNet architecture, as discussed in the study by Ishikawa et al. (2020). They utilized digital signal processing (DSP) for preprocessing tasks. ResNet architecture proved beneficial in overcoming limitations encountered with Convolutional Neural Networks (CNNs), particularly the vanishing gradient problem. This challenge was effectively addressed by ResNet signature data, highlighting the importance of ensuring data accuracy and consistency in such applications.
Key Words: Security risks, forgery, manipulation, deep learning algorithms, VGG16.
1. INTRODUCTION The handwritten signature stands as a critical biometric trait used for identity verification across legal, financial, and administrative domains [1], [2]. Manual authentication processes can be both time-consuming and prone to errors. Recent advancements in deep learning and computer vision have opened up avenues for more accurate and efficient automated signature recognition systems. These systems hold potential applications in sectors such as banking, law enforcement, and governmental organizations. However, despite their promise, they encounter challenges, particularly in accurately detecting forged signatures due to variations in styles, pen pressure, and angles. To tackle this issue, recent research has turned to Convolutional Neural Networks (CNNs), achieving high levels of accuracy in signature recognition, reaching up to 98.8%, and forgery detection, up to 89% [3], [4]. Architectures like GoogLeNet’s Inception-v1 and Inception-v3, employing CNN models, have also shown promise, with validation rates of 83% and 75%,
© 2024, IRJET
|
Impact Factor value: 8.226
B. Rateria and Agarwal (2018) introduced a novel approach in their paper on handwritten signature authentication. They combined a traditional Convolutional Neural Network (CNN) with a Siamese neural network to authenticate handwritten signatures. Two configurations were employed for detecting handwritten signatures in their study. The first configuration acted as a feature extractor, crucial for discerning the authenticity of a signature. The second configuration functioned as a classifier, utilizing a Siamese neural network. This innovative setup involved the use of twin identical networks within the Siamese architecture, representing a pioneering effort in utilizing dual networks to extract features and distinguish between authentic and forged signatures.
|
ISO 9001:2008 Certified Journal
|
Page 1230