Research On Off-line Automatic Chinese Signature Authentication Method Based On And-Or Tree

Signature authentication is a common way to identify,automatic offline Chinese signature authentication has an advantage of low cost and the more objective results,although many sch olars have carried out research on this subject,and achieved better results,but the authentication performance is still need to improve.Therefore,it is significant to further study on this subject to improve the authentication performance.By the need of offline Chinese signature authentication,the related techniques of image preprocessing are presented,including image binarization,image skeleton extraction,image denoising,blank area processing,inclination correction and normalization etc.On this basis,the connectivity of the signature image regions is analyzed for the following feature extraction.The signature image is divided into regions after the preprocessing.The pseudoZernike moments are used to describe the shape features of the local regions of signature,and the corresponding feature vectors are constructed.Based on the fact that there are many different writing styles of one signer,the differences of corresponding local regions between the multiple signatures are measured by Euclidean distance,and the spectral clustering is used to analyze the difference of the signature regions writed by one signer,and the And-Or tree feature models of the signature are constructed.The problem of Chinese automatic offline signature authentication can be viewed as a classification problem.Considering the global feature and the local feature,the appropriate classification decision function is designed based on the signature's And-Or tree feature models during classifying,and a set of optimal parameters is obtained by the support vector machine.The training experiment and the authentication experiment are designed,and are carried out by using the collected dataset,the experiment results show that the proposed method is more effective than the other existing methods,and it has better performance on the FRR?FAR and EER.