SIGNETIX - Writer Independent Fake Signature Detection System

Signatures act as a distinct means of personal identification and are widely used for verifying the relevant documents and legal transactions. They serve as a exclusive identifier for individuals in various applications that includes banking, official documentation, and contractual agreements. However, signature forgery still remains a vital challenge, making trustable verification methods essential. A key challenge in this domain is developing a system that can generalize across different individuals without requiring the user-specific training. This paper proposes a writer-independent offline signature verification system using Siamese Convolutional Neural Networks(SCNNs). In contrast to writer-dependent models, which consider for training on every signature, a writer-independent technique aims to identify signatures as genuine or counterfeit without having prior knowledge of the signer's writing style. Signature verification can be categorized into two types: static (offline) and dynamic (online). Offline verification consists of authenticating a signature after it has been made, which can be inefficient and time consuming when dealing with a large number of documents. To address these drawbacks, online biometric verification methods such as fingerprint recognition and iris scanning have gained popularity among the industry. However, an offline signature verification system is proposed using Siamese Convolutional Neural Networks (SCNN) in this paper due to its strong capability to extract high-level features from images. Unlike traditional approaches, SCNN automatically identifies important features without human intervention, making it highly influential for image classification tasks. Moreover, SCNN preserves superior accuracy compared to other algorithms used for image-based predictions, making it a suitable choice for offline signature verification.