One Shot Verification of Handwritten Signatures using Siamese Networks

In the real world, banks have always required a handwritten signature while creating a new account, signing cheques or any other official document. But they never really verify if the signature provided is authentic or just a forgery trying to fraud the person and the bank. Not only banks documents but other official documents like those in courtrooms, hospitals, and property documents also need to authenticate the handwritten signatures. Our paper introduces a novel and efficient offline signature verification method that provides a quick, secure and reliable real-time technique to verify and authenticate the handwritten signatures on any official documents. The proposed technique looks to automate the entire system to ensure that proper validation is done and forgery is detected at the source. Currently, some systems have been introduced to automate the process of signature verification but they face multiple drawbacks that our proposed solution manages to overcome in an efficient manner. The existing systems often require large datasets, training and execution time and face accuracy issues. The following algorithm uses a one-shot learning approach to validate signatures. One-shot learning combined with the Siamese network ensures that with the help of only one signature (taken maybe at the time of original account creation), we can train the model and spot the differences to display whether the signature is authentic or forged with high accuracy. Our algorithm improves on the current model by enhancing its speed and usability. Our solution minimizes the time and energy required in the verification process and drastically lowers the human error that may occur. In this paper, we also compare our results with other techniques while providing an improved solution.