Based On Hmm-svm Hybrid Model Line Handwritten Signature Verification,

Handwritten signature verification process is the use of unique personal writing rules to carry out authentication of a biological behavior characteristics of authentication methods.As a recognized identity authentication technology,it is widely used in the financial,securities,e-commerce and e-government and other fields because of its convenient,reliable and easily accepted by people,etc.So,it has important research value.In the current mainstream method of handwritten signature verification,maybe elastic matching method is faster but its recognition rate is not high;the recognition rate of HMM method maybe will be higher,but it only consider the role of positive training samples without considering the impact of negative training samples,which greatly limits discriminating ability of HMM;Although the neural network can be carried out self-learning according to the representative samples and have good robustness and adaptability,its structure is difficult to determine because it is difficult to determine hidden nodes.However,SVM method can gain higher recognition rate,but it also exists some limits because the samples of existence of cross-aliasing can not be accurately adjudged.Above the methods,HMM and SVM are very popular since a few years.HMM models is more suitable for dealing with continuous signals,but it requires substantial training data to estimate probability distribution,then SVM only requires a few training samples to gain better classification effect.Therefore,if full use HMM method's better modeling capabilities and SVM's higher classification ability,without too much training samples we can get a better verification results .Currently ,HMM-SVM hybrid model in face recognition,speaker verification and other fields has been verified and obtains some results.But most of the HMM-SVM hybrid models transform the output of SVM into probabilities,which is seen as the output probability of each hidden state in HMM,so that each state in HMM must correspond to an SVM. At this point if the number of states is too large,it will be bound to slow down the training speed.And when the positive training samples and the negative samples exist cross-aliasing phenomenon,especially for the characteristics data of the large samples,SVM can not accurately identify them.This article uses HMM-SVM+Sigmoid hybrid model for handwritten signature verification.It uses HMM to compress the characteistics data of the large samples and uses multi-dimensional probability vector of HMM as the input vector of SVM.As for the problem of cross-aliasing,SVM can not accurately identify them.by introducing Sigmoid function to transform the classification results to probability output,it can effectively solve the samples uncertainty caused by the cross-aliasing,and further improves the performance of signature verification.Finally,some experiments with this method are made on the SVC2004 Signature database in the Microsoft Visual C++6.0 development platforms.The result indicates that the proposed method can get better performance such as lower ERR compared to HMM,SVM and HMM-SVM.