摘要
利用卷积神经网络可以对破损、模糊不清的文字进行有效识别。为了实现速度快、精度高等优点,通过卷积神经网络中的LeNet-5网络模型对手写汉字图像进行识别。首先,在模拟写字板中建立手写汉字的图像数据集,搭建并训练卷积神经网络模型保存图像特征;然后对输入的手写汉字图像进行模拟污染并采用7种滤波去噪方式;最后对加噪、滤波处理后的图像进行识别,对比不同滤波处理的准确性。实验结果可表明,该方法能高效、稳定地从有噪声图像中识别出文字,同时经高斯滤波与PCA滤波处理后的图像识别精确度更高。
Using convolutional neural networks to identify damaged and blurred characters is an obvious development trend of automatic and intelligent archaeological relics.In view of convolutional neural network′s advantages of fast recognition speed and high accuracy,this paper studied the utility of applying LeNet-5 network model of convolutional neural networks to recognition of handwritten Chinese character images.First an image dataset of handwritten Chinese characters is established in a simulated tablet,and a convolutional neural network model is built and trained to store image features.After simulating the contamination of input handwritten Chinese character images,seven filtering methods are used for denoising,and then the images treated with noise addition and filtering processing is recognized to compare accuracies of different filtering methods.Experimental results show that the proposed method can achieve efficient and stable recognition of characters,and the recognition accuracy after Gaussian filtering and PCA filtering is higher.
作者
张静娴
冷青轩
陈航
李素真
ZHANG Jingxian;LENG Qingxuan;CHEN Hang;LI Suzhen(School of Artificial Intelligence and Electrical Engineering,Guizhou Institute of Technology,Guiyang 550025,China)
出处
《电工技术》
2023年第24期69-73,共5页
Electric Engineering
关键词
卷积神经网络
图像污染
滤波去噪
图像识别
高斯滤波
convolutional neural network
image contamination
filter denoising
image recognition
Gaussian filtering