基于多维特征和候选项的易混手写英文识别  

Confusable handwritten English recognition based on multi-dimensional features and candidates

作　　者：付鹏斌 宋冬雪 杨惠荣 FU Peng-bin;SONG Dong-xue;YANG Hui-rong(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124

出　　处：《计算机工程与设计》2021年第11期3204-3210,共7页Computer Engineering and Design

基　　金：北京市自然科学基金项目(4153058)。

摘　　要：针对手写英文识别中易混字符的识别问题,提出一种结合多维特征和候选项以区分易混字符的识别方法。利用卷积神经网络(convolutional neural networks,CNN)对手写英文字符进行识别,根据初始字符识别信息确定易混字符的类别;利用多维特征,设计针对不同类别易混字符的识别规则;由易混字符和其相连字符组成候选项单词,结合语料库以及字符间构成关系,最终对易混字符进行识别判断。实验结果表明,该方法在解决了易混字符的识别问题后,识别手写英文字符的平均准确率达到98.67%,具有一定应用价值。Aiming at the recognition problem of confusable characters in handwritten English recognition,a recognition method combining multi-dimensional features and candidates was proposed to distinguish confusable characters.Convolutional neural networks(CNN)was used to recognize handwritten English characters,and the types of confusable characters were determined based on the initial character recognition information.Multi-dimensional features were used to design recognition rules for diffe-rent types of confusable characters.Candidate words were composed of confusable characters and their connected characters,combined with the corpus and the relationship between the characters,the confusable characters were recognized and judged.Experimental results show that the average recognition accuracy of handwritten English characters can reach 98.67%after solving the recognition problem of confusable characters using the proposed method,and it has a certain application value.

关 键 词：手写英文识别 易混字符 卷积神经网络 光学字符识别 字符特征 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]