Off-line Signature Verification

Signature is one of the biological features used abroad to verify identity, and plays an important role in lots of fields.As a way of personal identification, signature verification has a wide perspective in applications in electronic-bank, military affairs, electronic-business, communication, office automation and seeurity. So it's valuable for us to deal with the off-line signature verification research.Signature images being the objeets, preproeessing, feature exrtraction, verification and recognition are studied in depth in this paper. Our research work focus on the following aspects:1) Image filtering,binarization,thinning and handwriting repairing are studied.This part proposes an hybrid binarization algorithm for signature image witch Overcome the shortcomings of global thresholding and local threshold effectively.Our hybrid approach is robust to noise and makes connectivity of stroke better.2) A method for feature extraction, which is based on LPP, is adopted in the paper.Three different kinds of features are used as initial features, including shape, pseudo dynamic and texture feature. It avoids the shortage of using only one kind of fearutes which are not capable of reflecting slight difference.Then the LPP is used to reduce the dimension of the feature space in order to get more discriminant features. This method has the nonlinear peculiarity of keeping the data structure unchanged.3) A method based on optimal threshold for recognition is adopted.Weighted euclidean distance is used for verification and genetic algorithm is used to find out the optimal threshold for each writer adaptivly,which can overcome the drawbacks of traditional method of single threshold. The experiment result shows that the approach can effectively let down the FAR and FRR, and has encouraged the performance of the verification.4) A method for signature recognition, which is based on a multi-class classifier composed of two class SVM classifier with error correcting, is adopted in the paper.Error Correction code combined with SVM is used to compose a multi-class classifier.The recognition results is decided by the output of each SVM classifier. This classifier has error correcting ability because it uses the error control encode in communicational channel. Therefor this classifier has better classified effect than traditional classifiers.