针对在手写字符识别中由于书写习惯和风格的不同而造成的字符模式不稳定问题,提出了一种基于流形学习的手写体数字识别方法。在流形学习非监督的基础上引入了监督信息,从而保证高维到低维的映射在保留流形某些结构的同时也可进一步分离不同类别的流形。算法首先利用基于监督的局部线性嵌入(SLLE)对手写体数字图像进行字符特征的降维,然后再对降维后的特征进行分类识别。对MINST库中手写体数字数据库进行了实验,实验结果表明,利用SLLE降维以后的特征能够有效地区分字符,识别率可达到93.27％;由于具有较好的识别率,能够发现高维空间的低维嵌入流形。

In order to improve the instability of handwritten character pattern caused by different writing styles, a novel handwritten numeral character recognition approach based on manifold learning is proposed in this paper.Based on non-supervised manifold learning,a supervised information is induced to the algorithm to ensure the map from high dimension to low dimension to retain some manifold structures and also to seperate different kinds of manifolds. By proposed method,Supervised Locally Linear Embedding (SLLE) algorithm is used to reduce the dimensionality of input feature.Then, the reduced feature is classified by simple classifier.Finally,the proposed algorithm is tested on the characters in MINST character database. The experimental results demonstrate that the method can effectively improve the recognition rate to 93.27% and can provide a new approach to the research of handwritten numeral character recognition.