Offline Chinese Handwriting Signature Verification Via Stability Evaluation

With the rapid development of modern information technology of the 21st century,handwriting signature verification plays an important role in broad fields.In this thesis,to improve the verification accuracy of Chinese handwritten signature,we propose a hierarchical classification method for Chinese handwriting signature verification called ISHC-SV(Integrated Signature Stability and Hierarchical Classifying Method for Signature Verification),which combining Extreme Learning Machine(ELM)with Sparse Representation.The main research of this thesis are listed as follows.(1)First of all,the Chinese signature images are pre-processed,then many signature features including static features and pseudo-dynamic features are extracted.(2)In order to improve the robustness of the signature verification algorithm,we first propose the approach of applying the stability of Chinese handwriting signatures to identify whether the signature is genuine or forged.First of all,we introduce Statistical Shape Model(SSM)and inner-class signature variance distance dispersion for stability analyzing and quantifying.Secondly,all signatures are classified into the stable and unstable ones.(3)To figure out the problem of signature verification under different stability,all stable Chinese signature sets are used to train the signature verification classifier,and the unstable ones are clustered to carry out template matching.(4)In order to train a high performance classifier of signature verification,we propose a hierarchical classification method of combining the advantage of ELM and sparse representation classifier.Comparing our method with One-Class Support Vector Machine method which is published in Pattern Recognition in 2015,the experiment results show 3%increases the average accuracy of our method in low stability signature sets.In addition,the Chinese handwriting signature verification accuracy of the proposed ISHC-SV method achieves 95.53%,which is better than two state-of-art methods.