CNN Versus Conditional Generative Adversarial Network cGAN–Digital Signature Authentication: Case Study

In this research work, the author gathered digital signature images from students in real time appropriate for identifying the fake signature and classifying the images into real and forgery. Convolutional Neural Network based approach, GAN, ResNet and CGAN techniques were employed to extend the system which can authenticate and identify forged signatures among students. The features are extracted from real time signature images using Principle Component Analysis, the most popular multivariate statistical techniques. Moreover, feature selection has done for choosing subset of best and worst features randomly from signature images, fed signature image parameters into neural network for training the images. The performance of proposed model is evaluated in terms of metrics such as accuracy, F1-measure and precision. Our experimental outcomes reveal that CNN model achieved 100% accuracy outperforms in detecting fake digital signature signed by students helpful in ensuring authenticity, integrity of documents, assignments, preventing frauds as well.