Handwritten Signature Verification Based On The Most Stable Feature And Partition

Biometrics is an authentication method for identity authentication by measuring human physiological or behavioral characteristics.Biometrics-based authentication overcomes many of the shortcomings of traditional authentication methods and has been widely used in many areas.Compared with other biometrics,signatures are widely used in many common business areas(eg e-commerce),including online banking transactions,electronic payments,access control,etc,because of their ease of access,low cost of acquisition and easy acceptance.In the handwritten signature verification method,the parameter-based method is fast and is not affected by the local time offset,but the error rate is usually higher;Function-based methods have lower error rates,but are time-consuming;region-based approaches reduce processing time and provide opportunities to retrieve local information,but partitioning is more difficult to determine.This paper combines the function-based approach and the region-based approach for signature verification.The function-based method compares the signature's similarity by the cumulative difference of two time functions.Two time functions are established by dynamic programming or HMM.HMM chooses the Markov model to spend a lot of time,and the training model needs a lot of computation.Therefore,this paper uses dynamic programming to establish the relationship between two signatures.In this paper,the online handwritten signature verification was studied,the main contents are:Firstly,a method of length regularization of signatures based on the most stable dynamic characteristics is proposed.Because the stability of each dynamic feature generated by each user signature is different,the accuracy of the relationship is different when the two signatures is established with different dynamic characteristics,which leads to the different accuracy of signature verification.According to this feature,this paper presents the most stable dynamic feature selection method for each user.Secondly,An improved dynamic time regular algorithm is proposed.The DTW algorithm computes the distance between the dynamic characteristics of the signature,using the European distance,ignoring the characteristics of the signature time series itself,so that when the waveforms of the two feature sequences are translated,where a sequence of local mutations or the amplitude of the waveforms do not coincide,The DTW algorithm may explain the variability of the Y-axis by distorting the X-axis,so that the matching of the local points is not accurate,which brings a certain degree of error to the authentication.The improved DTW algorithm takes into account the characteristics of the self-shape of the signature,that is,the trend of the curve around each point.Thirdly,A method of partitioning based on dynamic signature of signature is proposed.A region-based approach can reduce processing time and take full advantage of the signature's local information.In this paper,we use the three dynamic characteristics of speed direction angle,velocity and pressure to partition the user's signature.The stability of the signature after partition is different in different regions,a more stable partition should be selected for signature verification.And calculate the weight of each partition,select the largest partition for the most stable partition,for the final signature verification.Fourthly,an online handwritten signature verification method based on Relief algorithm to extract combinatorial features is proposed.In this paper,the signature dyna mic characteristics and similarity calculation method to form a combined feature.There are many combined feature,and some combined features are similar to each other,and some are unique.The combined feature of weak ability to distinguish authenticity s ignatures should be removed.In this paper,we use the feature discrete function to rough the combination feature,and then use Relief algorithm to re-filter the combination of features.Finally,the KNN classifier is used to verify the signature.