Font Size: a A A

Research On On-Line Hand-written Signature Verification Algorithm

Posted on:2008-04-29Degree:DoctorType:Dissertation
Country:ChinaCandidate:F J LuanFull Text:PDF
GTID:1118360212497794Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
With the development of the computer and network technology, the storage and transmission of electronic information is widely used because of its celerity and convenience, this created a need for electronically verifying a person's identity and many researchers focus on it now. Signatures are particularly suitable for authentication because each person's signature is highly unique, especially when the dynamic properties of the signature are considered in addition to the static shape of the signature. Even if skilled forgers can accurately reproduce the shape of signatures, it is impossible for them to simultaneously reproduce the dynamic properties as well. Authentication by on-line handwritten signature is one of the most accepted authentication systems for its normal and customary way.Based on the study of Discrete Fourier Transform, Dynamic Time Warp and Hidden Markov Model, the mathematical model of the system was constructed and three algorithms of on-line handwritten signature verification were presented in this paper. The main contents are listed as following. 1. On-line handwritten signature verification system which is including data acquisition, signature preprocessing, feature extraction, comparison process and authentication was systematically analyzed.2. A set of mathematical model based on time series was established according to the characteristic of on-line handwritten signature verification.3. An algorithm of on-line handwritten signature verification based on frequency region analysis was presented. The primitive signature signals, like the coordinate (X, Y) and pressure, were carried out by FFT. Low-frequency amplitude in the frequency region was extracted and judged whether the signature is true or not by it. For the personal characteristics of different signatures of a person, the stability of the result of FFT was analyzed. Weight was added by stability according to the signature of each person. The experiment demonstrated that this algorithm is simple, rapid and efficient.4. Two DTW algorithms based on some especial points and divisions were designed. The goal of the DTW algorithm is to find the most optimal time alignment between the reference signature and the signature in question. The characteristic of this algorithm is that it can accept the distinction and capture individual features of the signature at the same time. According to the characteristic of the signature's shape, the especial points which are the curved regions in the shape were extracted from the signatures. The Euclidean distance between the coordinate (X, Y) and the pressure series of the especial points were applied to DTW, the speed of verification was quickened and the performance index was same as before. A signature was divided into some little strokes by the especial points, the Euclidean distance between two end points of every little stroke, in addition to the different length, last time and the angle of the little stroke in the two signatures, were computed for DTW , which improved the effect of the recognition.5. HMM is widely used because of its time sequential nature of on-line signature as well as its capability of modeling in probabilistic terms. HMM model which is trained by Baum-Welch algorithms is constructed. In the verification phase, the Forward-Backward algorithm is used. The Threshold used in the verification phase is a function of the minimum probability in training phase. If this probability (which is called the likelihood function) is high, the signature is accepted, otherwise it is rejected. Different kinds of HMM topology were tested and Viterbi algorithm was used to analyze the results.
Keywords/Search Tags:On-line Handwritten Signature Verification, Pattern Recognition, Fast Fourier Transform (FFT), Dynamic Time Warping (DTW), Hidden Markov model (HMM)
PDF Full Text Request
Related items