Published Paper Details:

Signature verification is a critical component of document authentication and fraud detection systems, with applications ranging from financial transactions to legal agreements. Traditional methods for verifying signatures often rely on manual inspection and feature engineering, making them susceptible to errors and time-consuming. In response to these challenges, this research presents an innovative approach that leverages deep learning techniques, specifically Siamese and Convolutional Neural Networks (CNNs), to automate and enhance the accuracy of signature verification. The Siamese network architecture is employed to compute the similarity between two signature images, allowing for a precise determination of their authenticity. This network comprises identical subnetworks that share weights and are trained to extract discriminative features from input signatures. The Euclidean distance between the learned embeddings produced by these subnetworks serves as a measure of similarity, enabling reliable signature verification. Complementing the Siamese network, a CNN is employed to extract hierarchical features from signature images. This deep learning architecture includes convolutional and pooling layers, enabling automatic feature extraction. Extracted features are subsequently fed into fully connected layers for classification, providing an additional layer of verification. Experimental evaluations were conducted on two distinct datasets: the Cedar dataset, comprising genuine and forged signatures, and the ICDAR dataset for classification tasks. The results demonstrate the efficacy of our proposed approach. The Siamese network achieved an impressive accuracy of X% on the Cedar dataset, effectively distinguishing between genuine and forged signatures. Additionally, the CNN attained a validation accuracy of Y% on the ICDAR dataset, showcasing its ability to classify signatures as genuine or forged. In conclusion, this research offers a compelling solution for automated signature verification by combining Siamese and Convolutional Neural Networks. By addressing the limitations of traditional methods, our approach contributes to the field of document authentication, with the potential to enhance security and efficiency in various domains.