摘 要
神经网络模式识别方法是近几年兴起的模式识别领域的一个新的研究方向。由于神经网络的高速并行处理、分布存贮信息等特性符合人类视觉系统的基本工作原则.具有很强自学习性 、组织性、容错性 、高度非线性、高度鲁棒性、联想记忆功能和推理意识功能。本文首先介绍了神经网络的基本概念,特别是BP神经网络的原理和基本算法。同时介绍了模式识别的各种方法,并进一步了解了字符识别技术的基本理论。主要讨论用人工神经网络方法对英文字符的识别,将标准印刷体英文字符进行二值化处理,用矩阵表示,从而进行识别。同时考虑了噪声干扰或者是非线性因素的存在。分别用理想样本和混有噪声的样本训练神经网络并绘出误差曲线进行比较。再者通过设置回调函数利用GUI界面开发字符识别系统,使其直观显示。
关键词 神经网络;字符识别
Abstract
Neural networks method is a new method in recent pattern recognition; it provides a new means to Character Recognition. It has the advantage over some traditional technology: Good tolerance, Classification capability, Parallel processing capacity and self-learning ability. Thus, the neural network identification is a good choice.
The passage introduces all kinds of methods of character recognition, mainly discuss using Artificial Neural Network method to recognize character, and consider the existence of noise disturbance or nonlinear factor, so the network has certain fault-tolerance capability, and finally use MATLAB to simulate character recognition.
The study focuses on the standard Print English character recognition. Union pattern recognition and artificial neural network algorithm for a large number of related, especially feed-forward neural network model which is the most useful in target identification at present and the BP algorithm, and making use of GUI platform to develop the character recognition system in the end.