【作者】 邓集杰；

【导师】 刘铁根；

【作者基本信息】 天津大学 ， 光学工程， 2010， 博士

【摘要】 目前国内绝大多数银行还是采用传统的人眼识别方法检验支票印鉴真伪,这种方法识别在准确度和速度方面已经难以满足当前金融系统的需求。近年来出现了基于PC机的印鉴自动鉴别系统,但存在设备体积大、成本高、实时性差等突出缺点,未得到大面积推广应用。随着嵌入式系统的不断发展,基于嵌入式系统设计的支票印鉴真伪自动鉴别系统的研究已经开展,但目前的研究主要集中在处理和识别算法上,并未将检测方式及快速性考虑在内。随着国内支票开始异地流通和个人支票的出现,支票的流通数量急剧增加,因此,在已有单张支票印鉴自动识别系统的基础上,研究集连续采集、识别于一体的支票印鉴快速检测的方法及关键技术具有重要意义。本课题旨在研究支票印鉴快速检测方法的关键技术,以实现支票印鉴图像的快速处理为目标,针对支票印鉴快速检测的特点,研究了支票印鉴快速检测时图像的校正、印鉴图像的分割及印鉴特征提取和识别等关键技术,基于此,设计并实现了能够实现支票印鉴快速检测的原理样机。主要研究工作如下:1、实现了支票印鉴图像的快速校正。在分析图像采集过程中畸变成因的基础上,建立了畸变和复原模型。通过错位拼接实现了图像构建畸变校正;采用两点法校正由传感器的不一致性所造成的灰度畸变;将因行进过程缺陷而产生的几何畸变分为纵向畸变和横向畸变进行分别校正,通过扫描边界与标准行进方向的偏差进行横向畸变校正,通过比较相邻两行间像素差异来确定并消去差异行的方法进行纵向畸变校正。实验结果表明,校正后的图像可以满足后续处理要求。2、提出基于图像背景分析的颜色分割方法。将颜色校正系数引入印鉴图像分割和二值化算法中,该方法将背景底纹颜色与图像的二值化方法结合起来。通过对不同颜色底纹背景下印鉴二值化图像质量的评估,求出各种颜色底纹背景图像的最佳颜色校正系数。在图像分割处理时,直接根据背景底纹的颜色调出最适当的颜色校正系数对图像进行颜色校正,将经过颜色校正的图像再进行颜色分割,获得印鉴的二值化图像。该方法能够在很短的时间内获得除去背景底纹且保留印鉴最多文字细节的二值化图像。3、研究了基于SURF算法的印鉴识别方法。在提取印鉴图像的局部SURF特征的基础上,基于欧式距离进行待识别印鉴和预留印鉴特征点之间的匹配和配准,利用印鉴笔画边缘差异实现印鉴真伪识别。实验表明,印鉴识别算法对相同印鉴的识别准确率在97%以上,不同印鉴的识别准确率在94%以上,识别时间不超过70 ms。4、设计并实现了支票印鉴快速检测系统硬件。建立了双处理器快速支票印鉴检测系统数据处理部分的模型,分析了SOC系统中数据位宽与总线位宽的关系,提出了数据位宽向总线位宽的转化方法。设计和选择了系统的照明光源、镜头、图像传感器、模数转换芯片、驱动信息芯片等,在此基础建立了快速支票图像采集的测试系统。更多还原

【Abstract】 At present, most banks still identify the authenticity of the check seal through the traditional method of human eyes. It is difficulty to satisfy the need of current financial system in accuracy and speed. In recent years appeared PC-based sole check seal automatic identification system, but existing equipment big volume, the high cost and the non-real-time faults. Now, along with the development of embedded system based on embedded system design, the sole check seal authenticity automatic identification system research has been paid more attention to. But the system adopt sole check recognition, therefore detection efficiency is low. With domestic check began to different circulation and personal checks occurrence, check the circulation quantity also increased dramatically, therefore, the existing leaflet check seal automatic identification system on the basis of research set continuous sampling, recognition in one of the check seal fast detection methods and key technology are of great significance.This topic is to develop a rapid authenticity of batch seal inspection system of check seal. The aim of system is for high-speed processing seal image. On the basis of analysis of images treatment of check seal and algorithm detection and combination with the characters of the algorithm and the batch processing of check, rapid authenticity detection method was analyzed and the related design method from image data collection to process was proposed. And then, a dual-core high-speed check seal inspection system based on embedded system combination with check seal images collection, pro-process and authenticity automatic detection was realized in this paper. The main work is as follows:1、Realizing the checks seal images quickly calibration. On the basis of image acquisition process, it was Established that distortion and rehabilitation model. It was realize image splice correction through the dislocation joining. Two points methods was used to calibrate sensor inconsistency. It was used that two method to correct the Horizontal distortion and Longitudinal distortion. For Horizontal distortion, standard trajectory comparison method could be used to correct. For Longitudinal distortion, Through comparing the differences between two adjacent pixels rows to determine and delete redundance pixels rows to correct works. Experiment results showed, the check image after correction can satisfact the follow-up processing requirements. 2、It was proposed the color segmentation method based on image background analysis. The color calibration coefficient was introduced to seal imprint segment and binary algorithm. According to the different color background seal binary image quality evaluation, find out all sorts of color background image best color calibration coefficient. In the image segmentation processing, according to background colors, the most appropriate color calibration coefficient of image could be found directly and used it to color segment and seal binary algorithm. This method could in a very short time get remove background and retain the seal up the text details binary image.3、It was proposed the color segmentation method based on image background analysis. The color calibration coefficient was introduced to seal imprint segment and binary algorithm. According to the different color background seal binary image quality evaluation, find out all sorts of color background image best color calibration coefficient. In the image segmentation processing, according to background colors, the most appropriate color calibration coefficient of image could be found directly and used it to color segment and seal binary algorithm. This method could in a very short time get remove background and retain the seal up the text details binary image.4、Design and implement of hardware of check seal fast detection system. The model of double processor quick check seal detection system data processing parts was established. The relationship between data bits wide and bus bitwidth in SOC system was analyzed. The transformation method of data bits wide to the bus bitwidth bitwidth was proposed. Finally, the light sources, lens, image sensor, frequency-field chips, drive information chips of the system were designed and chosen, and then test system of quickly check image acquisition finished.更多还原