The Research Of Reducing False Reject Rate In Online Handwritten Signature Verification

With the development of network and communication technology and the increasing real and virtual space of human, we need status identification which can reach higher accuracy, security and practicability in our daily life. Then, in stead of traditional identification method biometrics becomes a new one.Among these biometric methods, an important one is handwritten signature verification which makes use of signature habit. Compared with off-line technology, on-line handwritten signature verification technology can obtain more useful characteristics of signature habit for us. Through specific handwritten input equipment, the shape, pressure and time distribution can be gathered. And then, by matching these characteristics real time signatures can be distinguished.Mathematic models in signature verification:One signature is represented by a series of sample points through gathered equipment. Each point is denoted by a vector (x, y, z, t) , in which x , y , z and tare the X-coordinates, Y-coordinates, pressure and time of the point respectively. Thus a handwritten signature can be viewed as a function with the independent variable t and dependent variables x, y and z .Generally the time span and the number of sample points of two signatures G (t) and F (t) are different. So we define a unitary transform D to solve this problem.Define the distance between the two signatures F(t) and G(t) asThen, the signature can be verified by judge the following inequation:Weight function based on stability analysis:During research we found that signature information included stable components which stood for the signature habit and the instable components which stood for stochastic information. The verification cannot be accurate without eliminating these stable components. Then the false reject rate (FRR) cannot be effectively controlled while the false accept rate (FAR) is lowered to 2ero.To reduce FRR, we designed weight function. During verification, a rational weight function can put more weight on the stable components represented signature habit and minish the bad influence of instable stochastic information.Denote the weight function with p(t), the formula (3) is amended asHere we design the weight function by stability analysis which can evaluate the stability on points, strokes or subsections of a stroke. In our experiment, we define the stability value in terms of the extent of the fluctuation of the signature and eachstroke. The stability value on the i th stroke s(i) isand the weight on the i th stroke isGive two kinds of weight function as follow:Cluster method based on priority value:Usually all the signatures of one man can be classified as several types. The signatures in different types seem more different and in the same type seem more similar. Then, we cluster the sample signatures of one man in signature library firstly. And on each cluster we define weight function respectively. As long as the input signature G and one of the signatures of the object man in the library satisfy the above equation (4), we regard it as a true signature.Based on the agglomerated hierarchy cluster method, we design our specific cluster method for signature verification.Optimize and renovate the signature library by the priority value:The human signature habit may change with the time went off. To track the new habit, we define the priority value. The higher the priority value means the newer the signature. Then, according to the priority value we can renovate the signature library, regulate the clusters and compute the more rational weight function.For the sample signature Ft, the priority value a(Fi) =...