基于机器视觉的文档与印鉴缺陷检测的方法与实现

发布时间：2018-01-10 19:07

本文关键词：基于机器视觉的文档与印鉴缺陷检测的方法与实现　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着机器视觉技术与自动化技术的发展,机器视觉技术已被广泛应用到各种缺陷检测中。本文利用机器视觉技术与图像处理领域相关知识,针对在线检测系统中,证书打印、盖章的缺陷问题进行检测。主要包括以下内容:(1)给出了一种基于边缘像素点旋转的文档倾斜角度的快速计算方法。该方法先将文档图像进行降采样处理,再对降采样的图像使用沈俊算子提取边缘。然后,采用由"粗"到"精"的方法对提取的边缘像素点进行旋转投影,并计算投影方差。由于方差最大时对应的旋转角度最可能是文档图像的倾斜角度,故据此来计算倾斜角。实验结果表明,使用该方法能够快速计算文档图像倾斜角度。(2)提出了一种基于直方图波峰分析的文档缺墨检测方法。该方法首先去除文字笔画周围灰度过渡区。然后,采用滑动窗口对图像进行遍历,计算滑动窗口内图像直方图,根据直方图波峰分布情况判断是否缺墨。最后,统计每列检测到的缺墨滑动窗口个数,若其大于一个阈值T,则该列为缺墨列。实验结果表明,该方法能够对缺墨文档图像做出准确判断。(3)提出了一种基于残差图的印鉴缺陷检测方法。该方法首先对印鉴进行定位,接着将待测印鉴与标准印鉴进行配准,然后计算配准后印鉴的残差图,再对残差图进行形态学滤波处理,最后对形态学滤波处理后的图像测量目标的周长与面积。根据周长与面积大小判断印鉴是否存在缺陷问题。在印鉴定位中,本文根据圆形印鉴的对称特征进行印鉴定位,该方法能快速定位印鉴位置且鲁棒性较高。在印鉴配准中,本文采用了 ORB算法提取待测印鉴与标准印鉴的特征点,对特征点进行匹配,再使用投票策略,计算待测印鉴与标准印鉴之间的偏转角,该方法能够有效地计算待测印鉴与标准印鉴之间的偏转角,实现配准。(4)实现了基于灰度直方图特征和SVM的文档缺墨区域检测。根据文本图像的特征,本文选择了灰度直方图特征,使用SVM分类器对滑动窗口内图像进行检测。实验结果表明,该方法能够有效判断窗口块是否缺墨。(5)最后本文介绍了一种面向计量检定测试中心的证书缺陷在线检测系统。主要介绍了其硬件系统与软件系统。
[Abstract]:With the development of machine vision technology and automation technology, machine vision technology has been widely applied to detect various defects. By using the machine vision technology and image processing knowledge, according to the online detection system, certificate printing, detect defects sealed. Mainly includes the following contents: (1) gives a fast calculate the edge pixel point of rotation of the document based on the tilt angle method. The down sampling processing of document image, and then use the Shen Jun operator to extract the edge image down sampling. Then, by using the "rough" to "fine" method of edge pixels to extract the rotation projection, and the projection variance. Because of the rotation angle corresponding to the maximum variance is most likely the tilt angle of the document image, so according to the calculated angle. The experimental results show that this method can quickly calculate the document image Tilt angle. (2) proposed a histogram peak analysis method based on document missing ink. This method firstly remove the strokes of characters around the gray transition zone. Then, using the sliding window to traverse the image, the image histogram is computed in the sliding window, according to the histogram peak distribution to determine whether tuppo ink. Finally, the statistics of each list of detected missing ink sliding window number, if it is greater than a threshold value T, the column is short of Molie. The experimental results show that this method can make accurate judgments on the ink document image. (3) proposed a seal defect detection method based on the residual graph. This method first carry on the localization the seal, then sample seal and seal standard registration, and then calculate the residual map seal registration, for residual image morphological filtering, finally measuring target on the perimeter of the processed image morphological filter The area and perimeter and area. According to whether the size of existing defects in the seal seal. The seal positioning, positioning according to the symmetrical characteristic of circle seal, the method can quickly locate the position of the seal and high robustness. In seal registration, this paper uses ORB algorithm to extract the feature points to be measured with the standard seal seal. To match the feature points, and then use the voting strategy, calculate the deflection angle between the seal and the seal of the standard to be measured, the method can effectively calculate the deflection angle between the measured and standard, seal seal to achieve registration. (4) the gray histogram and SVM document missing ink region detection according to the characteristics. The text image, this paper chooses the feature of gray histogram, sliding window image detection using SVM classifier. The experimental results show that this method can effectively block the window to determine whether the lack of ink (5). At last, this paper introduces a certificate defect online detection system oriented to metrological verification test center. It mainly introduces its hardware and software systems.

【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】