Algorithm Of Signature Verification Research Based On HMM/DTW TWO-Level Architecture

With the rapid developments of computer networks and communication technologies, information security issues face more and more challenges. It's an urgent and higher task in the modern social life,economic,politics and military as how to verify identification and ensure information security. The biometrics has a significant advantage over traditional authentication techniques and is attracting more and more scientific researchers'attention and be applied into reality .Identification based on On-Line handwriting is an important research area in the biometrics recognition. Compared with other biometric authentication methods, It is no-intrusive, convenient and more easily accepted by people. In this thesis, we use F-Tablet which is developed by Chinese Academy of Science collecting the signature's trajectory data as well as force information. Based these signature's sample data, we proposed a new algorithm which using HMM combined with DTW two–level architecture to verify authentication and get a very well result. The main content of the thesis consists of:Firstly, we reviewed the current state and background of the art in handwritten signature verification.Secondly, we proposed to use the two general signature segments method synthetically, and then invented a new signature segments match method based statistic principle which can effectively enhance the match accuracy in category.Thirdly, we put forward to a new algorithmic of signature verification based on HMM/DTW two-level architecture and carry out our signature verification experiments using trajectory, velocity and force information as the feature points ,and then give the requirement in the best result.Fourthly,we analsy the algorithm's certification and computation complexity then the related analysis result was given.Finally, we design an Electronic Medical Report Demo based on On-Line Handwriting Recognition, and make some research work on the application of handwriting recognition in EMR. And the future research directions were expected in the end.