基于人工神经网络的手写字母识别研究

发布时间：2018-05-02 20:08

本文选题：手写体识别 + 神经网络　；参考：《天津大学》2016年硕士论文

【摘要】：随着模式识别技术在信息科学中的广泛应用,手写文本识别也成为了现在的热点研究内容,在我们身边常用的就有手机的手写输入,车牌自动拍照提取,文档扫描输入等,都是要求识别图像中的文本。文本识别技术在很多需要自动识别信息的领域有很重要的理论意义和实用价值。本文研究神经网络方法用于手写文本识别。本文论述了手写文本识别的意义和发展现状,讨论了字符识别预处理的灰度化、图像去噪、二值化、归一化和字符细化等基本步骤。本文实验部分对手写48×48的彩色字符图像进行灰度化、二值化、归一化为16×16的黑白图像,字符特征向量提取用到了逐像素特征提取法。将提取的字符特征转换成神经网络的输入向量。选取20组手写样本训练BP神经网络,用另外20组样本测试神经网络的识别效果。本文采用不同训练函数的BP算法进行训练,比较识别效果和效率,最后采用带动量项的自适应学习率训练函数。本实验在MATLAB平台下实现,论文最后对实验结果进行了分析总结。本文研究表明,基于BP网络的手写文本识别的正确率较高,有一定抗干扰和噪声能力,将来可以进一步应用到实践中去。
[Abstract]:With the wide application of pattern recognition technology in information science, handwritten text recognition has become a hot research topic. All are required to recognize the text in the image. Text recognition technology has important theoretical significance and practical value in many fields which need automatic recognition information. In this paper, neural network method is studied for handwritten text recognition. This paper discusses the significance and development of handwritten text recognition, and discusses the basic steps of character recognition preprocessing, such as grayscale, image denoising, binarization, normalization and character thinning. In the experiment part of this paper, the color character image of handwritten 48 脳 48 is grayscale, binary, normalized to 16 脳 16 black and white image. The character feature vector extraction is based on pixel by pixel feature extraction method. The extracted character features are converted into input vectors of neural networks. Twenty groups of handwritten samples were selected to train BP neural network and the other 20 groups of samples were used to test the recognition effect of BP neural network. In this paper, BP algorithm with different training functions is used to compare the recognition effect and efficiency. Finally, the adaptive learning rate training function of the driving term is used. The experiment is implemented on MATLAB platform. Finally, the experimental results are analyzed and summarized. The research shows that the BP neural network based handwritten text recognition has a high accuracy, anti-jamming and noise ability, and can be further applied to practice in the future.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.43;TP183

【相似文献】