On-line signature verification based on PCA feature reduction and statistical analysis

Abstract

This paper presents a novel online signature verification method that uses PCA for dimensional-reduction of signature snapshot. The resulting vectors from PCA are submitted to a multilayer perceptron (MLP) neural network with EBP and sigmoid activation function. In the other hand, Dynamic features such as x, y coordinates, pressure, velocity, acceleration, pen down time, distance, altitude, azimuth and inclination angles, etc. are processed statistically. During enrollment, five reference signatures are captured from each user. One-way ANOVA is used to analyze relative X-Coordinates in 6 groups (5 reference group, 1 testing group). ANOVA test will be repeated for relative Y-Coordinates, pressure value, azimuth and inclination angles. Thus, the algorithm will fill up a vector of five distances (F-scores) between all the possible pairs of testing and reference vectors. The resulting vector is compared to a threshold vector. Our database includes 130 genuine signatures and 170 forgery signatures. Our verification system has achieved a false acceptance rate (FAR) of 2% and a false rejection rate (FRR) of 5%