Research On Feature Extraction Method Of Online Handwritten Signature Based On EMD And SVD

The development of information technology and the popularization of mobile equipment have greatly changed people's way of living and working with the continuous development of electronic commerce and mobile payment.But the issue of information security has becoming more and more serious.Therefore,a convenient and efficient user authentication tool is particularly urgent to put into practice.As a branch of biometric identification technology,online handwritten signature verification is an ideal means for personal authentication.Feature extraction is one of the key steps in the process of signature verification.It has a direct impact on the performance of authentication system.Empirical Mode Decomposition(EMD)is one of adaptive decomposition methods.Compared to wavelet transform,it not only has the advantage of multi-resolution,but also overcomes the difficulty of choosing wavelet bases.EMD has been widely used in signal denoising,fault diagnosis and other fields.In this dissertation,EMD is introduced into the signature feature extraction.An online handwritten signature feature extraction method based on EMD and Singular Value Decomposition(SVD)is proposed.The main works can be summarized as follows:1.An application for collecting signatures is designed and developed based on Android mobile phone.It is used to build an online signature database.2.An adaptive feature extraction method based on EMD and SVD is designed.It utilizes two dimensional coordinate information of these signatures to derive six time series data,such as velocity,acceleration,and so on.With the help of these time series,we get the initial characteristic matrix composed of intrinsic mode function obtained by EMD.Then the singular value vector is obtained by the singular value decomposition.The energy value of the singular values is used as a characteristic component of online signature sample.And then the feature vector is gotten by extracting the characteristic components of these six time series.The result of experiment shows that the inter-personal variability of the proposed method is clearly obvious.3.A signature verification model based on Support Vector Data Description(SVDD)is established to verify the validity of this proposed method.A small amount of the real signatures(we only take 6 real signatures)are used as the training samples in this dissertation.It has achieved a good recognition result.The average of false reject rate is 3.38%.Compared with the feature extraction method based on wavelet analysis,the proposed method has a certain advantage.