【作者】 朱均超；

【导师】 刘铁根；

【作者基本信息】 天津大学 ， 光学工程， 2007， 博士

【摘要】 印鉴作为标识身份的重要依据得到了广泛应用,特别是用于金融票据上,作为其法律有效性的最重要特征。目前印鉴的真伪辨别主要是通过人工“折角核印”,识别精度不能保证,伪造印鉴进行诈骗的犯罪活动时有发生,因此,印鉴真伪自动识别系统的研究具有重要意义。本文研究基于嵌入式系统技术的印鉴真伪检验系统,从系统的角度研究了金融票据圆形印鉴真伪识别方法,提出了从图像校正、预处理、特征提取到最终识别等一套完整的印鉴真伪识别算法,并设计、制作了基于高速DSP的嵌入式图像采集、处理系统,进行了圆形印鉴图像识别算法的实验和测试。主要内容包括:1、图像畸变校正研究。分析了图像采集过程中畸变的成因,建立了畸变和复原模型,根据该模型对畸变图像进行复原校正研究。对由于LED光源照度分布不均匀造成的图像非均匀性畸变,结合CMOS图像传感器的输出特性,提出了分段两点法对图像进行校正。对于图像的几何失真畸变,利用多项式拟合的网格模板图像,得到系统的几何畸变参数,然后利用该参数对图像进行几何校正和灰度插值,得到复原的图像。2、圆形印鉴ROI(感兴趣区域)提取算法研究和实现。以红色圆形印鉴为例,利用票据图像的彩色信息进行处理,把图像转换为灰度图像后,突出了原来的红色分量。再利用平滑卷积确定五角星上的一个像素A。以A为中心,半径为R(R大于印鉴半径)选择多个像素点。然后分别从各个象素点向A进行搜索,得到印鉴边框上多个像素点Bi,对其分析后保留标准差较小的m个点,忽略其他点。利用三点定圆法计算出印鉴的圆心和半径,提取到印鉴的ROI。3、计算印鉴偏转角度的研究与计算。把圆形印鉴ROI从圆心向其外边框进行投影,得到一维特征数据,在不同角度下分别计算SS(Sample Seal)和MS(Model Seal)投影之间的相关性,相似度最大时的相对角度就是所求偏转角度。实验结果表明,其定位精度≤0 .5°。4、圆形印鉴的真伪识别算法研究。在对SS和MS进行位置和角度精确配准的基础上,检测圆形印鉴的区域边缘。确定SS和MS之间的边缘对应关系后,采用改进Hausdorff距离计算其相似性,将其作为印鉴的真伪识别特征。为减小印鉴的质量对识别结果的影响,把边缘对应关系作为印鉴质量测度,采用神经网络方法进行综合分析、判别。实验表明,其识别率大于95%。5、算法在基于DSP的嵌入式系统中的实验与测试。采用高速的DSP作为主处理器,设计、制作了嵌入式圆形印鉴图像采集、处理系统,编写了实验和测试软件,对识别算法在DSP中进行实验和测试。实验结果表明,该算法可以在基于DSP的嵌入式系统中有效的进行圆形印鉴真伪识别,识别速度为1秒左右。在验证算法有效性的同时,也分析了算法各环节的处理速度,为算法的进一步优化和改进提供了依据。更多还原

【Abstract】 Seal plays an essential role in identifying entity status. Particularly, as a usual method, it proves law effect of financial notes stamped with seal. In most cases, the recognition of seal is completed with folding seal in diagonal and aligning manually. The accuracy of recognition result depends on man-made decision and grift crimes always happen with counterfeit financial seal, which are the reasons we manage to reaserch another unique method to solve this problem. Hence, A seal verification system is studied in this paper, based on embedded system technology. The recognition method for circle seal of finance notes is discovered from the view of system, and a series of recognition algorithms from image adjustment, preprocess, feature extraction to estimation are proposed. Based on high-speed DSP, the embedded image acquiring and processing system is designed and fabricated. Experimentally, the recognition algorithms is carried out.The followings are the main research contents:1. Image aberration adjustment. The factors leading to aberration are analyzed, and a model of aberration and reconstruction is established and is used for reconstruction adjustment of aberrated image. A section two points image adjustment method is proposed to solve the nonuniform aberration resulting from luminance distribution of LED light source, combining with the output character of CMOS image sensor. As far as geometric aberration is concerned, mesh template image obtained with polynomial fitting is used for geometric aberration parameter determination of system and then the parameters are used in geometric adjustment and gray interpolation for image reconstruction.2. Research and implement of ROI (Region of Interesting) extraction algorithm of circle seal. The red seal is selected as research sample. This seal is processed by color information of note image. For example, the color image is transferred to gray image to bring red component of original image into prominence firstly. Then a special pixel of pentagram is obtained with flatness convolution and defined as A. With the pixel A as center, multiple searches having direction to A are carried out until eight pixels in seal frame are obtained, these pixels are defined as Bi. Among of them, the pixels having smaller standard deviation will be kept while the others ignored. The circle center and radius of seal are calculated with three point determination method of circle, and then the ROI is extracted.3. Reaserch and caculation of measurement of seal orientation deflexion. Through the projection of seal ROI from its center to outer frame, a group of one dimension feature data is obtained. The correlation is calculated, between SS (Sample Seal and Model Seal) under different orientations. And then, we can determine the orientation with maximize the correlation. The experiment results demonstrate that the precision is≤0 .5°4. Authenticity recognition arithmetic of circle seal. The edge features of seal are adopted as recognition features, on basis of position and orientation precise registration of SS and MS. The corresponding relation, between edges of SS and MS, is determined and then similarity measure is calculated with improved Hausdorff distance. In order to alleviate the effect of seal quality on edge similarity, the corresponding relation is used as the measure of seal quality and artificial neural network is adopted in analyzing and estimation. Experiment shows that the recognition probability is up to 95%.5. Experiments and evaluation of recognition algorithm in DSP embeded system. This image acquirement and process system based on high speed DSP is designed and fabricated. Then the function measurement and performance of algorithm are carried out, which validate the system effectivity. The time consuming of each section of algorithm is analyzed, which provides the cornerstone for further optimum and improvement.更多还原