Writer-independent signature verification based on feature extraction fusion

Abstract

In this work, with the aim of separating the genuine and forgery samples of the signature, we developed an automatic handwritten signature verification (AH-SV) based on the informative feature. The large intra-class variation is an important challenge among specific writer samples in SV. Thus, the proposed method is focused on feature extraction fusion (FEF) to increase the accuracy and efficiency of the system. We use features such as extracting intersection points based on neighboring intersection pixels relative to the central pixel to achieve the most informative feature, which can be regarded as a model for estimating the transitional probabilities of the signature strokes. Discriminative features extraction for SV relies on the domain of graphology and texture description. Therefore, we extracted parametric features and local binary pattern (LBP) from the signature images, and then we applied fusion using canonical correlation analysis (CCA) to improve the discriminative features. Hence, the measurement of the proximity between genuine and forgery signatures can be computed by their corresponding feature vectors. We used MCYT, GPDSsynthetic and CEDAR datasets, with the classification of writer-independent structure on k-nearest neighbour (K-NN) classifier. Moreover, in the experiment phase, we selected random number of samples in datasets. The experimental results show that the accuracy in the CEDAR dataset has achieved the best value. Also, the results of the average error rate, false acceptance rate, and false rejection rate criteria have improved compared to well-known methods.