基于概率神经网络的手写体数字特征提取  

An Improved Method for Feature Extraction of Handwritten Digits Based on Probabilistic Neural Network

作　　者：李慧莹 胡西川[1] LI Hui-ying;HU Xi-chuan(College of Information Engineering,Shanghai Maritime University,Shanghai 201306)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《现代计算机》2018年第10期59-63,共5页Modern Computer

摘　　要：在脱机手写体数字识别项目中,重点研究在结构特征提取阶段,针对于二值字符图像的行列,使用统计该行或该列白线条出现的次数的方法,对比传统的统计该行或该列白像素出现的总量的方法,并且相应地在预处理阶段中采用"纵向拉伸"的字符填充方法,实验结果表明可使识别率提高近10.04%。该方法对手写体数字的结构表达更准确。实验使用MNIST手写数据库作为样本来源,利用概率神经网络进行分类识别,以识别率作为评价指标。In offline handwritten digit recognition projects, and it focuses on the structural feature extraction stage, a method of counting the number of the white pixels in the line or column for the binary character image is compared with the traditional method of counting the total amount of the white pixels, adopts the method of longitudinal stretching for character enlargement in the preprocessing stage accordingly, The experimental results showed that the recognition rate can be improved by 10.04%, this method is more accurate in the structure of handwritten digits. The experiment uses MNIST handwritten database as the sample, by utilizing the probabilistic neural network to classify, and the recognition rate is used as the evaluation index.

关 键 词：手写体数字 纵向拉伸 结构特征 概率神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.4[自动化与计算机技术—控制科学与工程]