基于模版对比的手写体数字识别神经网络模型

针对现有手写体数字识别神经网络模型的不足，提出基于模版对比的改进方法。建立8×12像素的手写体数字0～9的标准模版，则模版中每个数字与其他数字之间存在一定的像素差异，以此作为标准模版差异值。由于书写存在不确定性，采用在一定范围内随机增大或减小标准模版差异值的方法来构建神经网络模型的训练样本、检验样本与测试样本。在遵循建模基本原则和步骤的情况下，建立了泛化能力较好的手写体数字识别的神经网络模型。实验表明：该方法建模便捷、实用性好，测试样本的正确识别率达99.6%以上。

Artificial Neural Network model for handwritten digit recognition based on template matching

The shortcomings of the models，established in before，for Handwritten Digit Recognition（HDR） using Artificial Neural Network（ANN） are analyzed.A new method based on template matching is thus put forward.A standard template of handwritten digits 0～9 with 8×12 pels is created in this paper.The standard difference among one digit with the others is existed and picked up.Due to the uncertainty of handwritten digits，the training data，the verification data and testing data are then generated by adding a random number in a certain range to the standard differential value between the handwritten digit and the digits of standard template.The ANN model for HDR based on matching the template with good generalization ability is then established obeying to modeling principles and steps in this paper.The case study shows that the established model has good practicability.The accuracy of testing data is more than 99.6%.