Research On Off-Line Uyghur Signature Recognition Technology

Signature is one of the behavioral characteristics of personal identification thchnology.Handwritten signature identification have a wide application prospect in the field of financial,commercial, communications, office automation,and credit cards.The research on signaturerecognition technology have a great significance for practical applications and scientific andtechnological progress.The main work of this paper is divided into three aspects such as, preprocessing, featureextraction and selection, classification judgment. First consider the pre-processing stage processthe noising information in the signature image of the background.Second Because of thedifference the size of the signature and the location of signature on the paper, the signature imagesare normalized. Finally, consider the characteristics of feature extraction and refinement process ofthe signature image.In the feature extraction part, each Uyghur handwritten signature image is segmented intoseveral sub images with pyramid resolution to the16-dementional directional features, andthree-dementional base line and its shifted feature sare extracted in higher resolution layer, while32-dementional local central point feature and128-dementional edge of grid features wereextracted in the lower resolution layer.According to the data characteristics of different kinds of features, it is respectively classifiedto use K-NN classifier, Euclidean distance and chi-square distance based measure methods afterextracting effective features. It is comparative analyzed the effect of is Euclidean distance andchi-square distance based measure methods to the recognition rates, and the best measure methodfor the different fatures are confirmed. The obtained highest recognition rates separately is96%,95%,22.5%and90.50%via the classification experiments used four types of differentfeatures mentioned above. In order to further improve the recognition rate, the different featuresare combined together.in the result,the recognition rate of the combination feature of directionalfeature and the local center feature is rosed to98.50%,is the best state in this paper.