Abstract: A novel handwritten numeral recognition based on texture classification is presented.A handwritten numeral image is normalized to a fixed size and rotated by random angle degrees to create images which are used to form a texture image，then the handwritten numeral recognition is implemented by texture classification.The main frequency centers along multi-directions are extracted to form feature vector for a texture image and the minimum distance classifier is employed to classify the textures.Experiments show encouraging results.

Key words: handwritten numerals recognition, feature extraction, Hilbert-Huang transform, empirical mode decomposition

摘要： 提出了一种新的手写数字识别方法，通过将一幅规范化手写数字图像做任意旋转和简单排列，形成纹理图像，将手写数字识别问题转换为纹理识别问题。然后提取纹理图像在不同方法的主频中心作为特征向量，用最小距离分类器进行分类。实验表明，该方法不仅具有高的识别率和低的特征维数，而且具有旋转、伸缩和平移不变性。

关键词: 手写数字识别, 特征提取, Hilbert-Huang变换, 经验模式分解（EMD）