Handwritten signature forgery detection using Deep Neural Network
Manish Bag, Rajashree Dash, Deepak Pattnayak, Amisa Mohanty, Isha Dash · 2023
A common biometric for establishing a person's identity in document forensics is verification of handwritten signatures. Signature plays an important role in banking, financial, commercial and so on. However, issues associated with signatures are numerous since any two signatures may look very similar with little to no differences written by the same person. Despite the enormous research efforts, offline signature verification is still difficult, especially when attempting to differentiate between competent forgeries and authentic signatures because the visual differences between the two can sometimes be less noticeable than between two authentic ones. Due to the expanding usage of digital signatures and electronic documents, there has been an increase in the demand for effective and precise signature forgery detection systems (SFDS). In order to effectively combat handwritten signature forgery, this study intends to create an automated system utilizing a deep neural network (DNN) especially a simple and efficient Convolutional Neural Network (CNN) that can determine if a given signature is real or faked. The model will take signature image as input and extract important features through different layers of CNN that will helpful in distinguishing genuine and forged signatures. The model performance is also accessed with different image resizing techniques and optimizers used in training of CNN. The model's performance in terms of a set of assessment metrics, such as accuracy, precision, recall, and Fl-score are provided in this paper after extensive tests are conducted across a huge dataset of handwritten signatures including both genuine and various types of forged signatures. Additionally, the model's performance is also contrasted with a cutting-edge deep learning model. The simulation results conclusively demonstrate the appropriateness of the suggested methodology for automatically detecting forgery in handwritten signature images.