DEEP LEARNING-BASED SIGNATURE VERIFICATION

DEEP LEARNING

Every person has a distinctive signature that is mostly used for personal identification and the authentication of significant papers or legal transactions. Static and dynamic signature verification are the two options available. the process of using static (off-line) verification to confirm a paper or electronic file Dynamic (online) verification occurs while the signature is being formed on a digital tablet or other such device, whereas static (offline) verification occurs after the signature has been created. For many documents, offline signature verification is inefficient and slow. We have seen a rise in online biometric personal verification such as fingerprints, eye scans, etc. to overcome the limitations of offline signature verification. In this project, Python was used to build a CNN model for offline signatures. Following training and validation, the model's testing accuracy was 99.70%.Signature verification is a crucial duty in many different applications, such as banking, legal documents, and forensic investigation. Traditional methods of signature authentication rely on manually crafted classifiers and extracted properties, which usually struggle with scaling, handling changes in writing styles, and forgery detection. Recent developments in deep learning have shown promise in a variety of pattern recognition applications, including signature verification. This paper suggests a convolutional neural network (CNN) and recurrent neural network (RNN)-based deep learning method for signature verification. CNNs are employed in the suggested patch to automatically extract distinguishing qualities from input signature images. To capture the temporal correlations and sequential information included in the signature, the learned features are subsequently input into an RNN. The RNN generates a verification score that indicates whether the signature is likely real or fake.An extensive collection of real and fake signature photos is needed to train the deep learning model. The collection contains ground truth labels that define the veracity of each signature. By applying supervised learning to optimise a loss function that penalises misclassifications, the model is produced.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.