Online Handwriting Signature Verification Based On HMM/WNN Hybrid Model

With the rapid development of information technology and networking, information security is very important, and the technology of personal identification is practically significant. Among the various techniques of personal identification, handwriting signature verification has the characteristics of convenience, uniqueness and stability, and therefore is adopted more and more broadly.Focusing on handwriting signature recognition, this thesis proposes a novel method combining both advantages of hidden Markov model (HMM) and wavelet neural networks (WNN). The main contributions are as the following:Firstly, the general procedure of on-line handwritten signature recognition is introduced. Focusing on the step of feature extraction, the approaches of extracting optimal features from the signatures are discussed and then evaluated through the experiments (Chapter2).Secondly, an algorithm of HMM for online handwriting recognition is introduced and a signature recognition system is built. The experimental results show that the numbers of the Gaussian kernels and the states are the two key factors for accurate recognition. A higher recognition accuracy can be obtained by selecting the appropriate two key numbers and the optimal features (Chapter3).Thirdly, integrating WNN into the above HMM based online handwriting recognition system, a new recognition method is proposed. HMM is good at analyzing time series while the expertise of WNN is classification. Combining the advantages of WNN and HMM, the proposed method can increase the recognition accuracy (Chapter4.2and4.3).Finally, to verify the feasibility and the applicability of the proposed method, the evaluation experiments are performed on two different handwriting databases. The experimental results show that the recognition accuracy of the proposed method outperforms that of the HMM algorithm(Chapter4.4).