Study On Key Technology Of Off-line Chinese Signature Verification System

With the technology security becoming more and more important, automatic handwriting signature verification is to be one of the most significant ways of biometric identification as an important branch of biometric verification. Generally speaking, it aims at endowing computers with the ability to the detection of one or more categories of true or forged signatures. Comparing with this, off-line signature verification is not enough optimistic, depending its particular technique to some extent. However, off-line signature verification is not only easily accepted by more people, but also has less dependence on some electronic tools comparing with the forward, which will provide much more wide market and development.This context just makes a further research mainly on three aspects:preprocessing, feature extraction, feature selection and classification. The main contents can be described as follows:Firstly, in order to reduce effects on stable characteristics issuing by the preprocessing of signature as far as possible. As the traditional ways display:rotation, size normalization, displacement normalization and some other operations. Because some signature characteristics are changed during the procedure, this context just takes action in three aspects:binarization, discarding parse noise and extracting forward image, which have a significant effect on improving the capability of signature verification system.Secondly, considering of signatures'complexity, two important characteristics of signature are adopted. They are invariant moment and high gray. Although amount of calculation is large to some extent, the invariant moment mainly reflects the figurate characteristic and has excellent characteristics in size invariance, rotation invariance and displacement invariance which avail to the extraction of signatures'stability features. High gray characteristic also possesses the same traits above, which provide a significant reference for feature extraction.Thirdly, a more effective algorithm for signatures'feature selection is proposed. Aiming at gathering more stable signature, we takes an action in filtering some too distorted signatures in the same classification, which can improve the stability of feature to most extent, mostly in building a more stable and more efficient system of signature verification.For the Forth, an effective GA-BP algorithm in classification is put forward. Because the standard BP algorithm not only has a low convergence rate, it is more easily trapped in local minimum during its training. Generally speaking, GA acquires an excellent global search capability, while GA-BP-APARTING algorithm shows best in local approach. They can not avoid dropping in local minimum. The consequent GA-BP-NESTING algorithm is proposed. Because of new weights from BP network, GA algorithm based on these values, adding the nest circle. The algorithm can avoid the dropping in local minimum in theory.In the end, as we know from experiment analysis, after preprocessing, feature extraction and feature selection, the GA-BP-NESTING algorithm is adopted in exercising signature verification system. It has attained higher accurate and more capability than those of BP algorithm and GA-BP-APARTING algorithm.