Advanced Deep Learning Algorithm for Offline Signature Fraud Detection
Padma Priya K, A. P., Balaji B · 2024
One of the biometric methods for personal identification that is commonly employed is signature verification. The method of verifying a signature in many business contexts, including bank check payments, relies on a human scrutiny of a single known sample. Few attempts have been undertaken to conduct the verification based on a single reference sample, despite the fact that automatic signature verification has been the subject of much research. This paper proposes an offline method for verifying handwritten signatures using deep convolutional neural networks (DNNs), an explainable deep learning technique, and a feature extraction methodology based on BRISK. The advantages of Binary Robust Invariant Scalable Key-points (BRISK) over the Fast and SURF algorithms include much simpler computations, the use of distance rather than Euclidean distance, and faster execution times. We train our system on the open source document analysis and recognition (ICDAR)2011 Sigcomp dataset which helps us determine if a questioned signature is authentic or a fake. Every sample that we utilize for testing comes from a new author whose signature are absent from the training. The testing dataset has a higher accuracy based on the experimental outcomes