The Research Of Off-line Handwritten Signature Recognition Based On Bayesian Theory

With the development of computer technology and information technology, the communication among human beings becomes more and more convenient. However, the problems of information security are becoming even more prominent. Therefore, it is unquestionable that real-time and accurate individual certification would come into being. As a means of well-known individual certification, handwritten signature has made up to the basic defect of traditional identity certification concerning about password, IC card and so on; compared to other biological features, it also characterizes with its non-oblivious, natural and shared features. so handwritten signature receives wide use. Owing to its huge potentials in application and extra difficulties, there are great theoretical and practical significances on studying off-line handwritten signature recognition by using computer. The main research work of the full text is summarized as the following:The thesis aims at the images of Chinese handwriting signature. first of all, I reviewed the overview of the signature certification technology and the significance of the thesis, discussed the current situation, hot argument and difficulties of the recognition of handwritten signature, and summed up the studying progression in the field both at home and abroad, and raised deficiencies and made rooms for its progressing, and then around four basic research contents of pattern recognition:data acquisition, preprocessing, feature extraction and selection as well as classification and recognition, generally discussed to several key technologies involved in off-line Chinese handwritten signature recognition. In the theory of signature images, the thesis has done some meaningful jobs in the two fields:first, basing on discussing the basic theory of Bayesian, I presented an Bayesian classification algorithm of improving mean value, and made the correct identification rate to 89%, FRR reducing to 7.18%; second, combining model and evidence theory, I proposed a KNN classification algorithm of a variable K value, which overcomes the uncertainty of the value of K and obtains a higher recognition rate. Finally, through the experiments, we carried out classification and recognition on 20 groups of signature data, and proved the validity of the method.