Handwritten Signature Detection and Verification Using CNN

A signature which is manually handwritten is one of the best ways used for authentication and verification of legal documents in any official transaction or work. The real identification of the signature is very sensitive task because various factors like time, age, and environment and emotional factor is responsible for its variation from the original one. For the above mentioned reason, this report presents a CNN architecture-based deep learning model for detection and verification of signatures. As per experimental analysis, Google Net model has been used for the feature extraction purpose which works by transfer value calculation. Further for classification purpose, the classification layer was retrained with previous results using back propagation technique along with the idea of transfer-based learning. The classification layer of the deep learning model was retrained with 22 classes of signature image dataset with each class consisting of 16 signatures. After training, the model was evaluated with a testing dataset of 30% signatures from each class. The accuracy of model gets to be 98%, and validation accuracy become 88% with loss of 8%