A comparative study of the k nearest neighbour, threshold and neural network classifiers for handwritten signature verification using an enhanced directional PDF

A neural network approach is proposed to build the first stage of an automatic handwritten signature verification system that will eliminate random and simple forgeries rapidly. The directional probability density function was used as a global shape factor, and its discriminatory power was enhanced by reducing its cardinality. The choice of the best pretreatment was made by means of a k nearest neighbour classifier. This study has shown that the cardinality of the PDF can be reduced by a factor of ten while doubling its discriminatory power. The backpropagation model was retained to build the neural network classifier. An experimental protocol was used to find the best configuration of the BPN classifier whose performance was compared on the same database and with the same decision rule (without rejection criteria), to those of the kNN and threshold classifiers. This comparison shows that the BPN classifier is clearly better than the T classifier, and compares favourably with the kNN classifier.