On-line Signature Verification Based On Support Vector Data Description

How to identify a person’s identity accurately and swiftly is an urgent requirement that must be addressed in today’s information society. With the development of computer and communication technology, Biometric has become an increasingly important method for personal authentication. As one main research areas of the biometric, handwritten signature verification has also got more and more attention and research widely. Comparing with other biometric methods, on-line handwritten signature verification is more safe, reliable and convenientIn this dissertation, the feature function method of online signature verification based on support vector description (SVDD) has been deeply and systematically studied relying on the handwriting information acquisition platform of the F_Tablet which can captures the three dimension writing forces.The main works of this dissertation can be summarized as follows:1) Put forward a method for the original data of the signature compression based on discrete wavelet transform, which is both for data compression and reduction of computation, but also eliminating the noise and improving the quality and reliability of the data.2) To solve the difficulty of how to choose the verification threshold in on-line signature verification, we proposed a novel method based on template matching approach and SVDD, in which a hypersphere contains all training samples is created in feature space and verification threshold is chosen automatically by SVDD. Investigate the details of the method and examine the validation with features extracted from3-axis force of pen-tip to writing tablet and the effectiveness of the proposed solution is demonstrated by the experiment.3) We develop a new class of support vector data description by incorporating an idea of dynamic time warping into the kernel function (SVDD-DTWK).An approach for on-line signature verification based on DTAK-SVDD which can carry out automatic choice of matching template and decision threshold is presented and the details of the approach is investigated. The validation is examined with features of3-axis force of pen-tip to writing tablet. Experimental results indicate the effectiveness of the proposed solution. Moreover, we put forward a new selection method of target sample for SVDD-DTWK and SVDD algorithm with outliers. The results of verification experiments show that the proposed method is feasible and effective.4) There can be a large number of features for on-line signature verification, however, not all these features are useful, and we propose a kernel based method for measuring the consistency and discriminative power of these features. Then, we study on the information fusion applied in the biometric and propose a multi-SVDD classifiers fusion algorithm to improve the accuracy of identity verification. Experimental results show the effectiveness of this algorithm.