基于多特征集成分类器的脱机满文识别方法

为提高脱机满文手写字体的识别率,提出了基于BP网络的多特征集成分类器识别方法.对扫描成图像的手写满文进行预处理,切分出满文字元;分别提取满文字元的投影特征、链码特征以及端点和交叉点特征,并对这三类特征及其相互组合进行分类识别；通过隐马尔科夫算法对识别结果进行后处理,进一步提高识别的精度.实验结果表明,集成分类器的识别率要比单个特征的识别率要高,同时集成分类器中的特征类别越多,识别效果越好.

Off-line Manchu character recognition based on multi-classifier ensemble with combination features

To improve the off-line Manchu handwritten character recognition rate,a method of recognition based on the multi-classifier of back propagation neural network ensemble with combination features is presented.Firstly,the preprocessing is performed to segment the Manchu character units aiming at Manchu character image.Secondly,it is implemented to recognize the projection feature,chain code one and begin and end point and cross point one of Manchu character unit and the combination features of these ones.Finally,the post processing of Manchu character recognition result is done by the method of hidden Markov model and the recognition rate further is improved.The result of the experiment shows that the recognition rate of the multi-classifier ensemble is higher than the single one and the more features,the better in the multi-classifier ensemble.