| 本文已被：浏览 4926次 下载 5736次 |
码上扫一扫！ |
|
|
| 基于主动判别函数的手写体识别 |
|
孙广玲1, 刘家锋1, 唐降龙1, 石大明2, 赵巍1
|
|
1.哈尔滨工业大学,计算机科学与技术学院,黑龙江,哈尔滨,150001;2.School of Computer Engineering, Nanyang technological University, Singapore
|
|
| 摘要: |
| 提出了一种新的基于主动判别函数的手写体识别方法.首先,提出了主子空间中基于统计特征的主动原型模板并给出了对应于待识模式的最优主动原型模板.然后,提出了主动判别函数,它是两个加权距离之和;其中一个是待识模式与主子空间的距离,另一个是待识模式与次子空间的距离.其次,作为主动判别函数的参数,主动原型模板的约束可应用最小分类错误准则进行优化,这一最优约束有助于提高主动判别函数的识别正确率.最后,在银行票据中使用的手写金融汉字样本库上进行实验.实验结果表明,主动判别函数在手写体识别方面是非常有前景的. |
| 关键词: 手写体识别 主动判别函数 主动原型模板 主成分分析 最小分类错误准则 |
| DOI： |
| 分类号: |
| 基金项目:Supported by the Foundation of Harbin City for the Intellectual, Heilongjiang Province of Chinaunder Grant No.2004AFXXJ053(哈尔滨市后备人才基金项目) |
|
| Active Discriminant Function for Handwriting Recognition |
|
SUN Guang-Ling,LIU Jia-Feng,TANG Xiang-Long,SHI Da-Ming,ZHAO Wei
|
| Abstract: |
| A novel recognition method called Active Discriminant Function (ADF) for handwriting recognition is presented. First, statistical feature based Active Prototype Model (APM) in the principal subspace is proposed and an optimal APM corresponding to an unknown pattern is obtained. Second, ADF that is a weighted summation of two distances is proposed. One measures the distance between an unknown pattern and the principal subspace; the other measures the distance between an unknown pattern and the minor subspace. Third, as parameters of ADF, constraints for APM are optimized by applying Minimum Classification Error (MCE) criterion. The optimal constraints help to improve recognition accuracy of ADF. Finally, experiments are conducted on handwritten financial Chinese characters used in bank bill, and empirical results demonstrate that ADF is fairly promising for handwriting recognition. |
| Key words: handwriting recognition active discriminant function active prototype model principal component analysis minimum classification error criterion |
