中文摘要：

为了能够综合利用隐马尔可夫模型（HMMs）分类器在分类过程中能够得到的多种信息，提出一种基于距离相似性度量对HMMs后验概率进行调整的方法，将样本相似性与HMMs后验概率有机地结合起来进行识别。在分类过程中，采用距离相似性度量来描述待识别样本与模式类标准样本间的相似性，然后采用归一化距离相似性度量对后验概率进行适当调整，最后用调整后的概率进行分类。实验结果表明：与标准的HMMs识别方法相比，改进后的方法能够在计算量增加很小的情况下，较好地改善系统的识别精度；系统性能的改善效率在1．1～6．5间。

英文摘要：

A new approach for character recognition is propoed, which combines the distance similarity and the posterior probability to enhance the performance of the classifier based on hidden markov Models（HMMs）. In the proposed method,each class is represented by an HMM and a prototype. After the feature extracting,the similarity between the unknown sample and a prototype is calculated and normalized. Then the normalized distance similarity is used to adjust the posterior probability output by the corresponding traditional HMM. The experiments on Chinese legal amount recognition show that the novel method can effectively improve the recognition accuracy, and the recognition speed declines little. Moreover, the extent of performance variation for the adjusted system is from 1.1 times to 6.5 times.