摘 要

Abstract II

第1章 绪论 1

1.1课题提出的背景和意义 1

1.2纸病检测领域的研究现状 1

1.3设计研究的主要内容及论文结构 3

1.3.1本文研究的内容 3

1.3.2本文结构安排 3

第2章 纸病检测核心技术介绍 4

2.1常见纸病介绍 4

2.2纸病检测流程 5

2.3图像平滑滤波去噪 5

2.3图像特征提取 6

2.3.1阈值分割 6

2.3.2边缘检测 7

2.4模式识别算法 8

2.5本章小结 10

第3章 纸病检测算法整体设计 11

3.1算法程序简介 11

3.2图像预处理滤波去噪 11

3.2.1滤波算法的原理及程序实现 11

3.2.2滤波仿真结果对比 12

3.3图像分割 13

3.3.1阈值分割 13

3.4特征提取 16

3.4.1形态特征 17

3.4.2灰度特征 17

3.4.3特征量的选取 17

3.4.4边缘提取 18

3.4.5 K近邻算法 20

第4章 实验结果及分析 22

4.1算法调试坏境介绍 22

4.2实验方案及结果分析 22

4.3实验总结 27

第5章 总结与展望 28

5.1全文总结 28

5.2展望 28

参考文献 30

附录一：图像识别源程序 31

附录二 迭代法求阈值 35

致谢 37

摘要

纸张作为日常生活中书写记录的必备之物，与人们学习生活息息相关，是当今社会不可或缺的产品之一，在我国轻工业占有重要的地位。随着社会经济的发展，我们生活中越来越离不开书籍等与纸张相联系的产品，自然纸张的需求量也越来越高，所以造纸企业的发展趋势即是纸张的相对生产量越来越高。自然而然的现代的企业纸车速度也就越来越快，纸张幅面也越来越宽，为了能够满足它的技术要求，对处理纸病的算法的快速简洁性要求越来越高，因此本文针对常见的三种纸病孔洞，裂痕与亮斑，在纸病检测的预处理，分割，特征提取中分析各种算法优缺点，旨在找寻一种简单，快速，实时有效的算法满足工业需求。具体做了以下几个方面的工作：

（1）本文首先介绍目前纸病检测中主流的算法核心技术，主要从滤波，分割与特征提取这几个方面去分析介绍，具体分析他们的优缺点，并在此基础上探寻适合本文的纸病识别算法。

（2）在前面算法核心技术研究介绍基础上，我们主要对特征提取做重点分析在图像预处理之后，我们要选择合适的特征进行纸病判断，本文从纸病的形态学和像素等级层次去分析考虑，发现孔洞与亮斑大多呈圆形或近似圆形，而裂痕都近似呈长条形，因此我们首先利用圆形度这个数学概念将它们划分为两类，然后从像素等级去考虑，发现它们的灰度有着差异，通过灰度特征就可以具体把它们区分开来。

(3）本文纸病算法的整体流程设计，我们首先用现有的纸病图片对获取的纸张进行预处理，根据现场噪声的类型选择合适的滤波算法，主要把各种滤波处理效果进行了仿真对比，从实验结果选择一个最佳的滤波方法即中值滤波，然后对纸病图像做分割处理，首先对常见分割方法的介绍，并针对本文的纸病进行试验仿真与结果比较，分析优缺点，从本文实际算法简洁性考虑，本文选择的是较简洁的阈值分割，在阈值的选择上，把传统的迭代法与opencv函数库自带的自适应阈值分割做了比较，并选择更佳的一个，最后就是特征提取，根据前面分析，选择圆形度与平均灰度双重特征去识别判断。

（4）对常见模式识别算法分析与介绍，并针对本文双特征识别的特点，考虑到要提高算法简洁性，选择了训练简单的K-近邻算法来完成对三种纸病的识别，并完成了对它们的快速分类。

关键字：中值滤波，阈值分割，图像处理，纸病检测，特征提取

Abstract

Paper as a daily life’s necessary things, It is closely related to today's society and it is one of the indispensable products in China's light industry . With the development of social economy, we live more and more inseparable from the books, the demand for natural paper is also getting higher and higher, so the development trend of papermaking enterprises is the relative production of paper more and more higher. The speed of the modern enterprise paper production is getting faster . the paper format is more and more wide in order to meet its technical requirements. Processing paper disease fast and simple are more and more important, so this article is aim to the advantages and disadvantages of various algorithms in the pretreatment, segmentation and feature extraction of paper disease detection. It aims to find a simple, fast and real-time algorithm to meet the industrial demand. Specifically to do the following aspects of the work:

  (1) This paper first introduces the current core technology of paper detection in the mainstream, mainly from the filter, segmentation and feature extraction to analyze these aspects, the specific analysis of their strengths and weaknesses, and on this basis to explore the paper Disease identification algorithm.

  (2) On the basis of the introduction of the core technology research in the previous algorithm, we mainly focus on the feature extraction. After image preprocessing, we should choose the appropriate characteristics to judge the paper. This paper, from the morphological and pixel level of the paper Analysis and found that the holes and bright spots are mostly round or nearly round, and cracks are similar to the long strip, so we first use the mathematical concept of circularity and it will be divided into two categories, and then from the pixel level to consider, we can found that their gray has a difference, through the gray features we can be specifically to distinguish them.

  (3) As to the overall process design of the paper-based algorithm, we first use the existing paper image picture to obtain the paper pretreatment, according to the type of field noise to select the appropriate filtering algorithm, the main filter effect of a variety of simulation is in contrast. The optimal filtering method is selected from the experimental results, that is, the median filter. And then the paper image is segmented. First, the introduction of the common segmentation method is carried out, and the experimental results are compared with the results. In this paper, the simplification of the threshold is chosen. In the choice of thresholds, the traditional iterative method is