Real Time Signature Forgery Detection Using Machine Learning

Every individual has their own signature, which is primarily used for personal identification and verification of vital papers or legal transactions. Even today, in many commercial instances, such as check payment, register office the signature verification process is still relied on a single known sample being reviewed by a human. The probability of two signatures made by the same person is very less. So, forgery detection becomes a huge task. The signatures that undergoing verification is not only difficult but also time consuming, especially when they are signed offline and no information about the procedure is available. To prevent the possibility of theft or fraud, a system that can distinguish between real and faked signatures is required. Before recognizing the signature is forged or not certain stages of pre-processing must be done for the images (signature) are acquired from the dataset since it is in raw form. The proposed work is based on off-line signature verification using deep learning model incorporating Convolution Neural Network (CNN) and novel method for extracting local features. This method can be used in various organizations where the number of individuals are confined so the model can train those before detecting any forged signatures in the future.