基于手写-印刷混合字符的口算题识别  

Recognition of handwritten-printed mixed mental arithmetic

作　　者：纪睿哲 程艳云[1] JI Rui-zhe;CHENG Yan-yun(College of Automation and College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210042,China)

机构地区：[1]南京邮电大学自动化学院人工智能学院,江苏南京210042

出　　处：《计算机工程与设计》2021年第11期3271-3276,共6页Computer Engineering and Design

摘　　要：针对小学生口算题中印刷体与手写体字符同时存在以及手写体字符不规范的问题,提出`一种基于编码器-解码器模型的整体识别方法。引入多分支改进的Densenet网络对图片进行特征提取,在此基础上引入基于多分支改进后的联合CTC-Attention模型的编码器-解码器模型进行处理以充分利用多分支特征。基于真实样本数量不足、分布不均的问题,提出一种样本生成方法。以实例验证该模型与样本生成方法的可行性。To solve the problem of both printed part and handwritten part in the mental arithmetic of primary school students and the problem of irregular handwritten characters,an overall recognition method based on the encoder-decoder model was proposed.A multi-branch Densenet to realize feature extraction was introduced.Based on that,an encoder-decoder model based on the improved joint CTC-Attention model was also introduced for processing to make full use of the multi brand features.Meanwhile,a sample generation method is purposed to solve the deficiency and the unevenness of the datasets.In the experimental part,the effectiveness of the model and sample generation method proposed was verified.

关 键 词：手写-印刷混合字符识别 口算题识别 密集卷积网络 长短期记忆网络 联合模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]