Off-line signature verification using Neural Networks

This paper proposes a signature verification system that can authenticate a signature to avoid forgery cases. In the real world environment, it is often very difficult for any verification system to handle a huge collection of data, and to detect the genuine signatures with relatively good accuracy. Consequently, some artificial intelligence technique are used that can learn from the huge data set, in its training phase and can respond accurately, in its application phase without consuming much storage memory space and computational time. In addition, it should also have the ability to continuously update its knowledge from real time experiences. One such adaptive machine learning technique called a Multi-Layered Neural Network Model (NN Model) is implemented for the purpose of this work. Initially, a huge set of data is generated by collecting the images of several genuine and forgery signatures. The quality of the images is improved by using image processing followed by further extracting certain unique standard statistical features in its feature extraction phase. This output is given as the input to the above proposed NN Model to further improve its decision making capabilities. The performance of the proposed model is evaluated by calculating the fault acceptance and rejection rates for a small set of data. Further possible developments of this model are also outlined.