Multi-Layer Perceptron Neural Network for an Offline Signature Verification System

Signature verification using neural networks is characterized by the use of pre-processing techniques such as normalization, morphological operations and median filtering. In this work, an effective method for offline signature verification system based on multi-layer perceptron (MLP) was proposed. A signature can be divided into five logically connected, basic aspects or layers which are learnt by a single set of weights. The system was built based on a four-hidden layer neural network. An accuracy of 82.5% was attained in recognizing genuine and forged signatures which outperformed the state-of-the-art techniques that incorporate feature selection and preprocessing operations.