Offline Signature Verification System using CNN Algorithm Combined with Histogram of Oriented Gradients

One way of stealing someone's identity is through their biometrics such as their signature, wherein it is a measurement of the characterization of an individual based on their behavioral specifications. One, however, can imitate and forge another person's signature which can result in fraud and a huge loss. Thus, this paper explored offline signature verification by using a feature extractor with a neural network that classifies the signatures as forged or genuine. The research aimed to compare the results to the previously made system that used Artificial Neural Network in classifying the signatures. The Histogram of Oriented Gradients (HoG) extracted features and Convolutional Neural Networks classified the images and it was found that removing PCA and Skeletonization as a preprocessing step was the optimal model for the system. The developed system using CNN with HoG achieved an accuracy of 98.07% without skeletonization and PCA.