Verification of genuine and forged offline signatures using Siamese Neural Network (SNN)

Deep learning plays very important role in almost all domains and application areas of AI like computer vision, biometrics, NLP, Healthcare etc. However, when there is lack of training data then it is difficult to train a model. Siamese is one of the popular network in deep learning and applications. This network architecture is composed of two or more identical sub-network component and also shares same weights among them. The core benefit of this network is, it can learn from one input image along with one target image. Further it is identified as one shot learning. Generally, to avert falsification and discriminate geunine as well as forged signature, Convolutional Neural Network (CNN) is highly researched. The proposed work represents Siamese neural network using Convolutional Neural Network as a subnetwork for the proposed system. In Siamese network embedding vector is generated and to make this vector more robust, we proposed to add some statistical measures to it, which are calculated on embedding vector itself. Lastly contrastive loss function is applied to resultant embedding vector. The proposed network surpass the state-of-the-art results in terms of accuracy, FAR and FRR.