神经网络在手写体数字识别中的应用

Applications of Neural Network for Handwritten Digit Recognition

本文采用前向、多层神经网络,BP学习算法对40个人的手写体数字进行了识别。识别过程分为四步:首先,用HP扫描仪把写在纸上的数字变成二值图像,接着对它进行分割,规整等预处理,变换成32×32点阵。然后提取特征,把点阵图像变成特征描述。最后,进行训练和识别。在拒识率为25％条件下,得到误识率为0.4％的识别结果,文中还分析和讨论了在实验中遇到的一些问题。

This paper uses a feedforward, multilayer neural network with back propagation learning algorithm to recognize handwritten digits written by 40 persons. First, a HP scanner converted the original digits images to binary images. Then, some additional preprocessing is performed to segment and normalize the digits, the binary images are scaled to a 32×32 pixel matrix. And the paper extract the feature to change the representation of a digit from a pixel matrix to a feature description Finally, it achieved a result of 0.4% error rate at 25% reject rate in the computer simulation. Some problems encountered in the experiment are also discussed at the end.