针对目前手写数字识别精度不高的问题,通过对手写数字图像的研究,提出了基于手写数字图像的空间、旋转、层次和结构特性的特征提取方法.该方法把手写数字的统计和结构特征结合起来,以特征提取方法为基础,利用LibSVM算法对手写数字特征进行了训练和识别.通过实验给出了各个参数的推荐值,利用推荐参数值,手写数字MNIST字体库的识别率高达99.3333％.实验结果表明了该算法在识别手写数字上的有效性和准确性.%Aimed at the low accuracy of handwritten numeral recognition at present, through the research of handwritten numerals image. A method of feature extraction based on handwritten numeral images space, rotation, level and structure characteristics is proposed. This method combined statistical and structural features of the handwritten numerals. Based feature extraction method, handwritten numerals characteristics are trained and recognized through LibSVM algorithm. Recommended values to each parameter are given by the experimentation, and the MNIST handwritten numerals recognition rate can reach 99. 3333% by this recommended values. The experimental result shows that this method can recognize the handwritten numeral effectively and accurately.
展开▼
