基于组合特征的手写体数字识别方法

Handwritten Digit Recognition Method Based on Combination Features

提出了一种新的手写体数字识别方法。首先利用核主分量分析技术提取数字图像的全局特征,然后利用独立分量分析技术提取数字图像的局部特征,分别选出部分局部特征向量与部分全局特征向量组合成数字的组合特征向量,然后利用支持向量机分类器进行识别。采用USPS字库进行测试,并与其他特征提取方法进行了比较,实验结果显示基于组合特征方法的识别率明显优于其他方法。

A new method is proposed for handwritten digit recognition. Firstly, we extract global features using Kernel Principal Component Analysis （KPCA） technique and extract local features using Independent Component Analysis （ICA） technique. We select some of the local features and the global features and combine them. Then we perform classification using the combination features. For validation of the method, we tested our method on the USPS database by using linear Support Vector Machine. Meanwhile, we compared performance of our method with that of PCA-based, KPCA-based and ICA-based methods. The experiment results indicate the performance of our method is superior to those of other methods.