An Automated Signature Verification Model using Artificial Neural Networks

This paper presents a classification method for identification of forged and genuine signatures using artificial neural networks (ANN) based on prominent feature extraction. The features are responsible for uniquely identifying a signature and classify them in an automated manner once the intra-variability of signatures is addressed through all possible variations of signatures of the same person. In this paper, model is developed in three steps: first, pre-processing is applied on acquired image, then different features extraction is performed and lastly ANN based signature image classification is done. Total seven features namely the axis, centroid, eccentricity, solidity, skewness, kurtosis and corner detection are extracted and stored for both training as well as testing image and then compared to produce the final output. The classification decision is based on a threshold value. Performance of the model is evaluated on the dataset created for 200 persons; each person signature is taken at different angles and sizes containing 24 signature samples of an individual person in different physical state. The twelve different set of genuine signatures have been selected along with the corresponding forged signatures for training and testing. The model consistently achieves an accuracy between 94% and 98% across all scenarios. The results demonstrate that a specific combination of features contributes significantly to maintaining a high level of classification accuracy.