改进的贝叶斯多分类器组合规则

多分类器组合是提高识别效果的一条有效途径。文中提出一种用于多分类器组合的改进贝叶斯规则，即首先通过对大量样本的统计获得有关每个分类器识别性能的先验知识，将其作为多分类器组合的依据。组合时对每个类设置不同的阈值，使组合效果得以改善，这些阈值可以通过训练获得。在数字识别中的应用结果表明，改进的贝叶斯规则可以使多分类器的组合结果识别率和置信度得到明显提高。

Improved Bayes Principle for Combination of Multiple Classifiers

The combination of multiple classifiers is one of the effective ways to improve the recognition performance. An improved Bayes principle is proposed in this paper for the combination of multiple classifiers. A prior knowledge of each classifier is obtained by the statistics for a great many of samples, which acts as the basis of the combination principle. Different thresholds gained by training are employed for different classes,which can yield better results. Its application to numeral recognition shows that the improved Bayes principle made the combination of multiple classifiers achieve good performance of high recognition rate and high reliability.