基于贝叶斯评判子的字体判断

支票中的待识别文字既可能是印刷体，也可能是手写体．由于印刷体与手写体文字的预处理方法和识别算法不同，因此准确判断文字的字体(手写体或印刷体)是获得高精度识别结果的关键技术之一．该文根据贝叶斯决策理论的最小错误率判决规则，提出了基于评判子的字体判断方法．利用贝叶斯评判子，该文还提出了一种可分性判据：评判子散度；同时还给出了一种估算评判子函数的方法．在无拒识情况下，对12158张实际银行支票的测试中，该方法的正确率为99．4％．

Classification of Machine-Printed and Handwritten Texts Based on the Bayesian Judge

Correct classification of machine printed and handwritten texts is a key problem of getting accurate recognition result. Combining with the development of financial document OCR system, this paper presents a machine printed and handwritten text classification method using Bayes Judge. Based on Bayes Judge, a new separability measure, Judge Divergence, is put forward. A method for evaluating the Bayes Judge function is also presented. Using 12,158 actual bills to test the above method,the accuracy rate is 99.4%,which verifies the effectiveness of the method.