Automated signature inspection and forgery detection utilizing VGG-16: a deep convolutional neural network

With the growing usage of handwritten signatures in legal documents and financial transactions, it is crucial to select an efficient technique for validating these signatures and eliminating forgeries that might cause considerable losses for clients. Offline signature authentication is yet challenging despite extensive study, notably when trying to separate expert forgeries from signature pools. This research aims to develop an automated system using VGG-16 a deep convolution neural network (DNN) to successfully tackle handwritten signature fraud. The automated system is developed separately by utilizing a pre-trained and a trainable VGG-16, whose performance is accessed over a widely used signature dataset comprising an equal number of authentic and fake signatures. In comparison to the pre-trained VGG-16 model, the trainable VGG-16 based model can yield a testing accuracy of 0.962, which is about 17.75% higher. The simulation outcomes decisively illustrate that the recommended methodology is suitable for automatically spotting fraud handwritten signature images.