基于神经网络与比例法的多文字识别  

Multi-character recognition based on the neural network and projective method

作　　者：林毅斌[1] 姚剑敏[1] 郭太良[1] 

机构地区：[1]福州大学物理与信息工程学院,福建福州350002

出　　处：《电视技术》2016年第9期112-116,共5页Video Engineering

基　　金：国家"863"重大专项(2013AA030601);福建省科技重大专项(2014HZ0003-1);福建省资助省属高校专项课题(JK2014003)

摘　　要：用传统方法进行文字识别时,需要花费时间分别去提取各个文字的区域。但是在讲究速度与效率的实时系统中,文字识别的速度是一个值得深思的问题。提出了一种多文字识别方法,其核心思想是将输入的文字当做一个整体进行处理,而不需要在单字区域提取上消耗过多时间,再利用ROI以及投影法进一步减少比例法特征提取算法的运算时间。在此基础上,采用BP神经网络进行样本训练与文字识别。实验结果表明,该方法能够有效地实现快速识别,识别四个字的所花费的时间为120 ms。h is time-consuming to use the traditional method in recognizing characters by collecting characters by every word re- gion. However, it is worth of digging deeper into the problem of recognition speed, regarding to the efficiency in the real-time sys- tem. A multi-characters recognition method is presented. It mainly focuses on recognizing characters as a whole batch instead of spending much time in picking up every word region. In addition, it can make further improvement in reducing the computing time of characteristic parameter by using ROI and projective method. On that basis, BP neural network is applied in training samples and recognizing characters. The experimental result shows that it is an efficient method of characters recognition, which takes 120 ms in recognizing four characters.

关 键 词：多文字识别 比例法 BP神经网络 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]