Abstract: Due to the diversity of character recognition, a character recognition model based on Bagging ensemble is presented, which solves recognition model’s preferences for certain character. Different datasets are formed by Bagging, and then base-classifier is constructed. Ensemble learning model is built by majority vote. Theoretic analysis and simulation result shows the model owns better classification accuracy than other classification methods.

Key words: Bagging, character recognition, ensemble learning, decision tree, Adaboost

摘要： 针对字符识别对象的多样性，提出了一种基于Bagging集成的字符识别模型，解决了识别模型对部分字符识别的偏好现象。采用Bagging采样策略形成不同的数据子集，在此基础上用决策树算法训练形成多个基分类器，用多数投票机制对基分类器预测结果集成输出。理论分析与仿真实验结果表明，所提模型相比其他分类方法具有更好的分类能力。

关键词: Bagging, 字符识别, 集成学习, 决策树, Adaboost