Handwritten Signature Identification and Fraud Detection using Deep Learning and Computer Vision

Secure authentication plays a major role in ensuring the identity of a person. Everyone has an individual and unique signature, which is used as their personal identification in all their legal transactions. Even though a lot of things got digitalized, traditional signatures are still used in a lot of places, such as check payments and government offices, and they still rely on a human manually verifying them. As the customer base keeps increasing, this will become a greater and greater problem in the near future. So, forgery detection plays a key role in reducing these kinds of overhead. Manual verification is not only difficult to check if two signatures are the same but also very time-consuming. Pandemic further made people do tasks digitally, which also included uploading their own signatures digitally. This increases the urgency of implementing a system to identify and verify the user’s signature. A lot of existing techniques are patented now, and few are less effective. This paper proposes a method to pre-process the signature to make verification simple as well as use methods like the Convolution Neural Network (CNN), Scale Invariant Feature Transform (SIFT), Oriented FAST and Robust BRIEF (ORB), and Mean Square Error (MSE) to identify forged signatures and compare the results obtained with various parameters.