【作者】 常家艳；

【导师】 徐全生；

【作者基本信息】 沈阳工业大学 ， 计算机技术， 2007， 硕士

【摘要】 印章识别一直是一个比较困难的课题，目前银行使用的方法是手工操作，将支票上的印章像与客户留下的原始印章像进行手工折角比对，其缺点是人为因素多、准确性差、工作效率低。而实现印章管理和识别的电子化、自动化、智能化，能提高印章识别的准确性和可靠性，可以有效防止利用假印章进行诈骗的犯罪活动。论文首先介绍了图象识别的基础知识和特征提取的相关理论作为论文的理论基础，之后完成了印章识别的总体设计，在印章预处理的基础上，重点研究了印鉴纹理的特征提取的设计及其算法的实现，以及运用神经网络的判别方法对所提取的印章特征进行判别，以识别印章的真伪。在本文中，分析了防伪印章的特点，从图像纹理和分析的角度出发，对印章图像纹理的特征提取与选择特征参数进行了分析比较，设计了一种基于印章纹理的印章识别系统。预处理阶段：利用中值滤波去除图像的噪声后，采用基于颜色的印章图像分割技术将印章图像从背景中分割出来，在此基础上进行二值化，并通过边缘轮廓的提取是要分割出包含印章区域的最小区域。在特征提取阶段：采用灰度共生矩阵的方法来对印章纹理进行熵、能量、惯性距、局部平稳性、相关性的特征提取。在模式匹配阶段：利用BP网络的思想设计了一个真正的BP神经网络，经过样本训练后，对印章图像进行真伪的验证。通过实验数据说明该方法可以正确提取印鉴纹理的纹理特征，在与神经元网络结合后其，通过对手头样本的测试与实验，除了个别模糊的样本外，能够鉴别大部分印章，在有限人工干预的情况下基本能够满足部分应用。该方法的算法简单，并经实验验证及数据表明，该方法是从一种新的角度进行了印章识别，提出了识别印章的新的途径。更多还原

【Abstract】 Seal Identification has been a comparatively difficult thesis. It is done by hand in the bank at present, which is checking on the seal that customers left with the original seal through the manual folding. Its shortcomings are many personal factors, bad accuracy and low efficiency. It can be improve for seal’s accuracy and reliability to carry out the seal administration and identification electronic, automated, intellectualized, which can prevent effectively making use of fake seal to commit a defrauding crime.Firstly, the paper has introduced the relevance theory of image identification and characteristic extract, which is the theory basis. Secondly, complete the total design of seal identification. Based on the seal pretreatment, we study the design and calculation of vein characteristic extract and distinguish the seal feature abstracted using the neural networks discriminant, which is to distinguish whether the seal is truth or false.In this paper, we analyse the characteristic of guarding against false seal, From the image vein and the analytical angle, we compare with the vein characteristic abstracts and the characteristic parameter chose and design a seal Identification system based on seal vein. Pretreatment stage: Make use of middle value wave filter the image noise, adopt carved technology based on the colour seal image to carve the seal image from the background. Base on this, we carry our the value two here and carve up out the minimal area containing seal area with border outline extract. Extract a stage in the characteristic: We extract the entropy, energy, inertia distance, part balance, correlativity on seal vein. Stage mating in the pattern: Make use of network thought BP to designe a real BP neural networks. Carry out truth or false verification on seal image after the sample trained.Our data suggested that the vein characteristic can be extracted exactly using this method. After linking the neural network, we experiment the samplet on hand, we can distinguish most seal except individual illegibility sample. The algorithm is simple and verified. The data indicate this method which is distinguishing seal from the new angle and have brought a new approach to distinguish seal. The method can prevent effectively making use of fake seal to commit a defrauding crime.更多还原