【作者】 徐鹏；

【导师】 宋蛰存；

【作者基本信息】 东北林业大学 ， 农业电气化与自动化， 2009， 硕士

【摘要】 手写体字符识别是多年来的研究热点,也是模式识别领域中最成功的应用之一,一般可以分为两类:联机手写字符识别和光学字符识别(OCR或称离线字符识别)。联机识别是通过数字转换器来获得用户实际书写时的笔尖位置,进而获得数据,而离线识别主要通过扫描仪和数码设备获得数据的输入。其中,离线手写体数字识别已经成为近年来研究的热点问题,在许多领域都有其应用潜力,例如信件分拣、财务报表、银行票据、传真文件阅读等。国内外学者在这方面做了大量的研究,方法主要包括非线性主分量分析PCA算法、隐马尔可夫模型、支持向量机等。但是由于手写体数字字体变化很大,传统的识别方法要达到高的识别率,有较大的困难。本文针对离线手写体数字识别过程的特点,建立了基于BP神经网络的规范手写体离线识别模型,并进行了训练和仿真研究。首先,通过图像处理技术对通过数码设备或扫描仪采集的手写数字信息进行预处理,其次,对处理的图像数字化信息进行特征提取,最后利用BP神经网络的自学习、抗噪声和并行计算,以及强大的非线性映射能力,建立了随机样本数据的分类器模型和预测模型,通过仿真验证了此模型的有效性。由于BP算法属于局部寻优算法,在网络的训练过程中易陷入局部极小点,而影响网络分类器的识别率,针对这一问题,本文采用了基于PSO算法优化BP人工神经网络模式识别方法,建立了基于PSO-BP神经网络的规范手写体离线识别模型,并进行了训练和仿真研究,通过非线性函数的极值优化和样本数据的分类仿真,比较了PSO优化BP算法分类器和单独BP算法分类器的分类效果,结果表明PSO优化后的分类器具有较高的预测精度和泛化能力。更多还原

【Abstract】 Handwritten character recognition are widely researched over the years, but also one of the most successful applications in the field of pattern recognition, generally it can be divided into two categories: on-line handwritten character recognition and optical character recognition (OCR or off-line character recognition). Online identification is to get the user at the time of the actual writing nib location through digital converter, then access to data, but the off-line identification mainly through scanners and digital input data. Among them, the off-line handwritten numeral recognition has become a hot issue in recent years, it has many potential applications research in many fields, such as sorting of letters, financial statements, bank notes, fax paper reading. A lot of researches are done by domestic and foreign scholars in this regards, its methods include principal component analysis of nonlinear PCA algorithm, hidden Markov models, and support vector machines. However, due to changes in handwritten font large figure, there is more difficult to achieve high recognition rate for the traditional identification methods.In this paper, according to characteristics of off-line handwritten numeral recognition the process, off-line handwriting recognition norms model is proposed based on BP neural network, and conducted training and simulation. First of all, through image processing technology for digital equipment or collection of hand-written scanner for pre-processing of digital information, and secondly, to deal with digital information of the image feature extraction, and finally using BP neural network self-study, anti-noise and parallel computing, as well as strong nonlinear mapping ability, set up a random sample of data models and the classifier prediction models, simulation results show the effectiveness of this model.Since the BP algorithm belongs to the local optimization algorithm, the training process in the network is easy to fall into local minimum points, and the impact of network classifier recognition rate, in view of the problem, in this paper, PSO algorithm based on BP artificial neural network pattern recognition method , based on PSO-BP neural network off-line handwriting recognition norms model, and training and simulation studies, through the nonlinear function of the optimization and simulation of the classification of the sample data to compare the PSO algorithm to optimize classification of BP and BP algorithm classifier separate classification results, results show that the optimized PSO classifier has higher prediction accuracy and generalization ability.更多还原