Signature Forgery Detection using Siamese-Convolutional Neural Network

The process of identifying the authenticity or falsity of a signature is referred to as signature forgery detection. In the fields of document forensics and biometrics, checking signatures is among the most challenging duties. At least one type of deformation is sufficient to cause a skilled forger to diverge from an authentic signature, which distinguishes this verification problem from others in that it requires the modeling of minute yet critical distinctions between authentic and forged signatures. To represent an offline writer-independent signature verification task, we implement a convolutional Siamese network in this paper. By aligning similar observations in close proximity, twin networks that utilize shared weights and are simulamese are able to acquire knowledge of a feature space. The reduction of the Euclidean distance between similar and dissimilar pairs and the maximization of the distance between dissimilar pairs is achieved by training the network using a pair of dissimilar and similar observations. On the list of the most challenging responsibilities in biometrics is offline signature verification. This verification problem, as opposed to others, necessitates the mathematical representation of subtle yet critical differentiations between genuine and counterfeit signatures. This phenomenon occurs because a skilled forger can easily reproduce an authentic signature in specific signature verification tasks that employ a convolutional Siamese network. Consisting of voting systems with shared weights, Siamese networks are twin networks. In order to facilitate model training, a dataset consisting of authenticated or counterfeit signatures that have been annotated may be employed. After receiving training, the model may be employed to predict the genuineness of an original signature.