【作者】 韩贺磊；

【导师】 李剑忠；

【作者基本信息】 大连理工大学 ， 工业工程， 2007， 硕士

【摘要】 纸币面额识别是自动存款机和自动售货机等许多金融设备一个必不可少的功能模块。近年来，随着我国国民经济的迅猛发展，国内对自动化纸币清分机等金融设备的需求也变得十分迫切。本文在广泛查阅国内外相关参考文献的基础上，针对我国人民币纸币自身的特征，设计了基于图像处理和神经网络技术的面额识别方法。本文方法在先期获得人民币纸币图像的基础上，由噪声滤除、图像倾斜校正、图像定位分割、特征提取和面额识别等几个主要部分组成。①倾斜校正是本文方法中比较关键的一个环节，在倾斜校正过程中，采用了直线拟合的方法计算出纸币边框上边界所在直线的斜率，从而计算出纸币图像的倾斜角度，然后再进行图像旋转，使图像位置归一化。②在图像定位分割中采用了灰度投影的方法，根据背景图像和目标图像的明显差异，从而准确定位出纸币的上下和左右边界，使目标图像和背景分离开来，然后对目标图像进行幅面归一化。③在纸币的特征提取过程中，根据纸币图像的特殊性，采用类似轴对称掩膜方法提取人民币图像特征，减少了输入神经网络的模式类别。④在纸币的识别过程中，本文充分利用神经网络的并行分布处理和高的容错性特点，将BP神经网络算法应用于纸币的面额识别，不仅能够识别新币和感染轻微噪声的纸币，而且对符合流通要求的残币、旧币仍具有很强的鲁棒性，较好地实现了人民币面额的准确识别。通过对由扫描仪获得的一定数量的人民币图像进行识别实验，在允许有少量拒识现象发生的情况下，本文方法的正确识别率可达100％，完全满足当前国内对人民币面额的自动识别要求，具有很好的实际应用前景。更多还原

【Abstract】 Identifying of banknote denomination is a necessary function module on all financial equipments. In recent years, with rapid development of China’s national economy, the demand of automatic banknote detaching-machine and other financial equipments becomes extremely urgent. On the basis of references about recognition of banknote denomination at abroad and home, contraposing to RMB self-characteristics, this thesis designs denomination recognition method based on image processing and neural network technology.On the basis of RMB banknote images obtained in advance, this method is composed of image noise filtering, declining image adjusting, image locating and segmentation, feature extraction, denomination recognition and several other components.①Declining image adjusting is a more crucial tache. In the course of adjusting declining image, this thesis puts forward a method of calculating the slope of the upper frame line with line fit, consequently, the declining angle is obtained, and then we can rotate the image based on the above angle.②In the process of image locating and segmentation, a gray projection method is adopted. According to the obvious differences between image background and objectives, we can accurately locate the border of target image to separate target image from background image, and then normalize the target image.③In the process of feature extraction, the characteristics are extracted with similar axis-symmetry masks according to the specialty of the RMB image.④In the process of denomination recognition, we adequately use the abilities of parallel processing and high fault tolerance for NN, apply BP algorithm to denomination identifying of banknote. This recognition method not only recognizes new and slightly polluted banknotes, but also for defected and old but still circulating banknote is strongly robust, too. Through experiments for a certain number of RMB images acquired by the scanner, on the condition of a small amount of rejection, this method can achieve 100 percent correct recognition rate. It can fully meet the current domestic automatic recognition requirements for RMB denomination and will have good prospects for the practical application.更多还原