Artificial Neural Network with Histogram Oriented Swerve Angle for Signature Verification

Offline signature verification is a challenging task, particular in distinguishing between genuine signatures and skilled forgery. How to distinguish features and how to calculate the similarity score are a key issue in signature verification. This paper proposes a comparison offline handwriting signature verification algorithms for solving skilled forgery problem by using Artificial Neural Network (ANN) with our proposed methodology as the Histogram Oriented Swerve Angle (HOSA). The main methodology were developed by the technique of skeletonization also known as the thinning process, which is an important step in the signature pre-processing. After preprocessing process, we use the histogram oriented swerve angle for determining the signature image feature extraction. The verification process is accessed with an artificial neural network (ANN) classifier. This paper applied ANN with a comparison 2 optimization methods between Stochastic Gradient Descent (SGD) and Levenberg-Marquardt (LM). The experiment was conducted with CEDAR datasets. The results showed that Skeletonization and Histogram Oriented Swerve Angle techniques can apply ANN with SGD and LM to distinguish between genuine signatures and skilled forgery signatures.