An intelligent network for offline signature verification using chain code

Abstract

It has been observed that every signature is distinctive, and that's why, the use of signatures as a biometric has been supported and implemented in various technologies. It is almost impossible for a person himself to repeat the same signature every time he signs. We proposed an intelligent system for off-line signature verification using chain-code. Dynamic features are not available, so, it becomes more difficult to achieve the goal. Chain-code is extracted locally and Feed Forward Back Propagation Neural Network used as a classifier. Chain-code is a simple directional feature, extracted from a thinned image of signature because contour based system acquires more memory. An intelligent network is proposed for training and classification. The results are compared with a very basic energy density method. Chain-code method is found very effective if number of samples available for training is limited, which is also practically feasible. © Springer-Verlag Berlin Heidelberg 2011.