Abstract: Sample-based similarity is an important decision factor for classifying. But in conventional HMMs framework, posterior probability is the only base for classifying. This paper presents a new approach to recognize off-line handwritten character by combining sample similarity and HMMs. In order to measure the similarity between samples, three types of distance are selected and tested. Then the normalized distance measure is employed to adjust the posterior probabilities of HMMs’ outputs. Experimental results confirm the performance of the proposed methods. The recognition rate of the new approach is higher than traditional HMMs, but the recognition speed declines little.

Key words: : Sample similarity, HMMs, single classifier, the adjustment of posterior probability

摘要： 样本相似性是两个样本是否属于同一类别的重要依据，而传统的隐马尔可夫建模（HMMs）方法仅根据后验概率进行分类。本文将二者结合起来，提出一种基于样本相似性的HMMs后验概率调整方法。在该方法中采用距离来描述样本间的相似性，利用规范化的距离相似性度量对后验概率进行适当的调整。由于在一个单分类器中充分利用了两种分类信息，因此将其用于脱机手写大写金额的识别过程中，取得了良好的效果：在识别精度提到的同时，识别速度变化很小。

关键词: 样本相似性, HMMs, 单分类器, 后验概率调整