基于深度学习的盲文识别方法

A Deep Learning Method for Braille Recognition

提出一种基于深度学习的盲文点字识别方法,利用深度模型——堆叠去噪自动编码器(Stack Denoising Auto Encoder,SDAE)解决盲文识别中特征的自动提取与降维等问题。在构建深度模型过程中,采用非监督贪婪逐层训练算法(Greedy Layer-Wise Unsupervised Learning Algorithm)初始化网络权重,使用反向传播算法优化网络参数。利用SDAE自动学习盲文点字图片特征,使用Softmax分类器进行识别。实验结果表明,本文所提方法较之传统方法,可以有效解决样本特征的自动学习与特征降维等问题,操作更为简易,并能获得满意的识别结果。

This paper mainly proposes a deep learning method, using Stacked Denoising AutoEncoder （ SDAE） to solve the prob-lems of automatic feature extraction and dimension reduction in Braille recognition. In the construction of a network with deep ar-chitecture, a feature extractor was trained with unsupervised greedy layer-wise training algorithm to initialize the weights for ex-tracting features from Braille images, and then a following classifier was set up for recognition. The experimental results show that comparing to traditional methods, the constructed network based on the deep learning method can easily recognize Braille images with satisfied performance. The deep learning model can effectively solve the Braille recognition problem in automatic feature ex-traction and dimension reduction with a reduced preprocessing.