图像处理与卷积神经网络相结合的脱机手写汉字识别方法  

Method for Offline Handwritten Chinese Character Recognition by Combining Image Processing with Convolutional Neural Network

作　　者：陈悦 黄寄洪 Chen Yue;Huang Jihong(School of Data Science and Software Engineering,Wuzhou University,Wuzhou 543002,Guangxi,China)

机构地区：[1]梧州学院大数据与软件工程学院,广西梧州543002

出　　处：《梧州学院学报》2024年第5期54-62,共9页Journal of Wuzhou University

基　　金：梧州学院校级科研项目(2023B006)。

摘　　要：为了解决传统脱机手写汉字识别方法对相似手写汉字识别率低的问题,提出了一种图像处理与卷积神经网络相结合的两阶段脱机手写汉字识别方法:第一阶段使用传统的卷积神经网络进行识别,第二阶段使用基于图像处理的差异辨别方法进行更加精确的二次识别。试验结果表明:将基于图像处理的差异辨别方法与卷积神经网络结合起来比单纯基于卷积神经网络的识别方法能够更好地识别相似汉字,从而可以提高总体手写汉字的识别率。此外,所提出的方法还表现出更稳定的识别效果,可以有效应对训练样本中存在错误标注的情况。In order to solve the problem of low recognition accuracy of similar handwritten Chinese characters by traditional offline handwritten Chinese character recognition methods:the present study proposes a two-stage offline handwritten Chinese character recognition method that combines image processing and convolutional neural network,the first stage uses the traditional convolutional neural network for recognition,and the second stage uses the differences discrimination method based on image processing for more accurate secondary recognition.The experimental results show that the differences discrimination method based on image processing combining with convolutional neural network has better performance than the recognition method based on convolutional neural network alone,and thus the overall handwritten Chinese character recognition accuracy can be improved.In addition,the proposed method shows more stable recognition and can effectively deal with the case that training set contains mislabeled samples.

关 键 词：脱机手写汉字识别 相似汉字 卷积神经网络 差异辨别方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]