基于轮廓和统计特征的手写体数字识别  

Handwritten Digit Recognition Based on Contour and Statistic Feature

作　　者：张重阳[1] 娄震[1] 杨静宇[1] 

机构地区：[1]南京理工大学计算机系,南京210094

出　　处：《计算机工程与应用》2004年第9期83-84,89,共3页Computer Engineering and Applications

摘　　要：提出了数字规范化模板特征,并利用这一特征与轮廓分段特征相结合对手写体数字进行识别。首先使用基于轮廓分段特征的分类器进行识别,通过提高拒识率获得高可靠性的分类结果。然后由基于数字规范化模板特征的分类器对前一级分类器的拒识样本分类。实验结果表明分别基于这两个特征的分类器在分类结果上具有较强的互补性。实验的数据为真实支票上采集的10000个手写体数字样本,该方法的识别率为98.06%。The normalized template feature is proposed.A method based on this feature and the segmented contour feature for the recognition of handwritten numerals is also developed.Firstly,the classifier,which is based on the segmented contour feature and high reliable,is used.Then,the classifier based on the normalized template feature is applied to the samples rejected by the former.Experimental results indicate that the correlation of these two features is small and the Combined method is efficient.Test samples are10000handwritten digits getting from the practical checks,the correct recognition rate is98.06%.

关 键 词：分类器组合 手写体数字识别 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]