Main Article Content

Abstract

A signature has become an important human attribute, which can represent personal information. Human signatures are widely used to authorize documents, both paper-based as well as electronic-based ones.  However, such authorization still poses various privacy issues, such as signature duplication and forgeries. These may not be easy to be addressed, particularly when involving many documents. Hence, advanced procedures are required to verify the signature authenticity. In this paper, we propose a new method for automatic signature verification based on the digitalized signature images. The method comprises successive image processing techniques, such as cropping, resizing, gray-scaling and thresholding. The binary images as the results of thresholding serve as the features of the signatures and are used to train a single layer Perceptron neural network. The experiment in this paper uses 42 digitalized signatures images, collected from two subjects. The obtained images are divided into the training and testing sets, in which the training and testing sets comprise 14 and 28 images, respectively. In the experiment, the proposed method produces the average training and testing accuracies of 100% and 98.85%, respectively. These indicate that the proposed method is reliable for practical applications.

Article Details