基于多维特征向量的图像检索技术的研究

VIP免费

3.0 牛悦 2024-11-19 4 4 1.38MB 55 页 15积分

侵权投诉

摘要

ABSTRACT

第一章绪论 ...................................................................................................... 1

§1.1 研究背景与意义....................................................................................1

§1.2 基于内容图像检索的系统结构............................................................3

§1.3 基于内容的图像的现状和发展情况....................................................4

§1.4 国内外典型的图像检索系统................................................................5

§1.5 基于内容的图像检索系统的框架和用到的关键技术........................7

§1.6 本文的研究内容和结构安排................................................................7

§1.6.1 论文研究内容............................................................................7

§1.6.2 论文创新点................................................................................7

§1.6.3 本论文组织结构........................................................................7

第二章基于颜色图像检索算法 .......................................................................... 9

§2.1 基于图像颜色特征检索算法简介........................................................9

§2.1.1 颜色直方图................................................................................9

§2.1.2 累计直方图..............................................................................10

§2.1.3 颜色布局算法..........................................................................10

§2.2 基于图像颜色特征度量算法..............................................................10

§2.3 小结......................................................................................................10

第三章基于纹理图像检索算法 ........................................................................ 12

§3.1 简介......................................................................................................12

§3.2 常用算法..............................................................................................12

§3.2.1Tamura 纹理特征......................................................................12

§3.2.2 游程长度特征提取..................................................................13

§3.2.3 灰度共生矩阵特征提取..........................................................14

§3.2.4 傅立叶纹理特征提取..............................................................16

§3.2.5 小波纹理特征提取..................................................................16

§3.3 多尺度特征融合..................................................................................18

§3.3.1 算法简介..................................................................................18

§3.3.2 试验仿真..................................................................................20

§3.4 小结......................................................................................................21

第四章基于形状图像检索算法 ........................................................................ 22

§4.1 简介......................................................................................................22

§4.2 常用算法简介......................................................................................22

§4.2.1 基本形状特征..........................................................................22

§4.2.2 图像边界连码表示法..............................................................23

§4.2.3 边缘距离直方图表示法..........................................................25

§4.2.4 边界矩方法..............................................................................26

§4.3 边缘点夹角直方图提取法.....................................................................27

§4.3.1 主要算法简介..........................................................................28

§4.3.2 相似性度量..............................................................................30

§4.3.3 算法性质..................................................................................31

§4.3.4 试验仿真..................................................................................31

§4.4 小结......................................................................................................32

第五章高维特征检索智能算法 ........................................................................ 33

§5.1 高维特征检索简介..............................................................................33

§5.2 高维特征检索智能算法主要步骤......................................................35

§5.2.1 图像预处理..............................................................................35

§5.2.2 提取图像边缘和图像主目标区域..........................................36

§5.2.3 提取图像颜色纹理和形状特征..............................................37

§5.2.4 优化特征向量..........................................................................37

§5.2.5 建立训练样本..........................................................................38

§5.2.6 根据训练样本集建立适应度函数矩阵..................................38

§5.3 遗传算法主要步骤..............................................................................39

§5.3.1 初始种群的建立......................................................................39

§5.3.2 遗传迭代..................................................................................40

§5.3.3 小结..........................................................................................40

第六章实验与结果 ............................................................................................ 41

6.1 量化评价方法..........................................................................................41

6.2 测试库的建立..........................................................................................41

6.3 实验数据分析..........................................................................................42

第七章总结展望 .................................................................................................. 45

7.1 小结..........................................................................................................45

7.2 展望..........................................................................................................45

参考文献 .................................................................................................................47

在读期间公开发表的论文和承担科研项目及取得成果.................................... 49

致谢 .........................................................................................................................50

摘要

近年来，随着多媒体技术、网络技术和数字化信息技术的不断发展和成熟，

全世界数字图像的数量也在飞速增长。如何快速而有效的寻找我们需要的数字图

像逐渐成为生活中不可或缺的一部分。早期基于文本的“字找图”图像检索技术，

存在很多缺点，耗时并且主观多义，故不能满足大规模图像检索的需求。于是人

们开始把基于内容的图像检索(CBIR)作为检索的重点，该检索方式是一种使用相

似性技术进行匹配的方法，用户可以提供一幅或者多幅样本图像作为查询图像，

使用算法自动抽取出这些图像的固有属性，如颜色、纹理、形状等特征，然后通

过提取的这些固有特征属性进行计算和比较，最后将候选图像按相似度大小进行

排序。CBIR 可以突破原来文本检索的局限，产生的效果也比较理想。

在 CBIR 中，图像特征的提取是最重要的环节。一方面，在提取特征向量时，

必须选择提取足够多的能够尽可能完全描述图像的特征；另一方面如何去除图像

检索过程中产生干扰的和冗余的信息，这也有助于减少检索时的复杂度，还能提

高图像检索的准确率。

随着 CBIR 的使用越来越广泛，人们对它的检索精度也提出了更高的要求。仅

仅靠单一的检索模式已不能满足人们的检索精度需求。特别当图像质量下降和区

域信息特征信息不明确时算法很难得到可靠的检索。每种检索方式都有它自己的

优点和不足。如何能建立一种全面的可以根据不同类型图像而智能化的选择满足

自己的特征提取检索方式是我们本文主要讨论的问题。

本文首先简要介绍了一些常用的颜色、纹理和形状检索的基本方法，并且针

对基于纹理的检索方法无法分辨图像高频信息和低频信息的弊端，提出了一种基

于小波分解的多尺度特征提取的方法，融合了图像高频域和低频域的特征，与原

来检索方法进行仿真比较，该方法可以提高检索精确性。

而后分析了基于形状图像检索严重依赖图像分割和边缘提取的弊端，提出了

一种直接对图像边缘曲线像素点进行特征提取的形状边缘点夹角直方图方法，该

方法大大降低了图像分割和边缘定位的不准确而带来的后续操作误差。而且易于

操作。

最后针对平均设置多维特征加权系数和手工设置多维特征加权系数特征越

来越不能满足基于内容图像检索精度的需要,提出了一种新的基于训练样本集聚

类的多维特征向量加权算法。该算法需要手工建立训练样本集，提取出每个元素

的颜色、纹理和形状等多维特征，然后使用遗传算法寻找特征向量集的最优加权

系数序列，最后使用该序列计算测试集的特征值进行检索。经实验证明，该算法

比其它高维特征向量组合方法有更高的精确性。

关键词：基于内容图像检索；小波分解；多尺度分析；多维特征提取；

形状边缘点夹角直方图；遗传算法.

III

ABSTRACT

In recent years, with the multimedia technology ,network technology and digital

information technology continues to evolve and be mature, the number of digital

images is also growing rapidly. How to seek digital images we need quickly and

effectively has become an indispensable part of life.For early text-based "search by

word" image retrieval, there are many shortcomings,time-consuming and subjective

meaning,so it can not meet the needs of large-scale image retrieval. Then people

started to regard content-based image retrieval (CBIR) as the focus means, a way using

similar technologies to match the images.User can provide one or pieces of sample

image as the query image, then automaticaly extract the inherent properties of these

images through some alogorithms, such as color, texture, shape and other

characteristics, and the size of the candidate images are sorted according to similarity

after calculated and compared. CBIR can overcome the limitations of the original text

retrieval,and the effect is realitively ideal.

The image feature extraction is the most important part in CBIR. On the one

hand,you must extract enough feature vectors to completely describe the image

features as much as possible; on the other hand, how to remove the interference and

redundant information of the image retrieval process is important, which not only helps

to reduce the search time complexity, but also improve the accuracy of image retrieval.

With more and more widespread use of CBIR, retrieval precision also is put

forward to higher requirements. Only a single crawl mode can not meet the needs of

retrieval precision. Especially when the image quality degrades and regional

characteristics of the information is not clear,it is difficult to obtain reliable search

results for algorothim. Each search method has its own advantages and

disadvantages,and how to build a comprehensive way,which can intelligently choose

their own search methods of feature extraction based images of different types,is the

problem discussed in this paper.

This article briefly describes some basic methods of retrieval using color, texture

and shape.For texture-based image retrieval methods can not distinguish between high

frequency and low frequency, a wavelet-based multi-scale feature extraction method is

put up,combining image characteristics of high-frequency domain and low-frequency

domain.By simulating with the original search methods, it proves this method can

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

15 积分 4人已下载

立即下载 VIP免费下载

摘要：

目录摘要ABSTRACT第一章绪论......................................................................................................1§1.1研究背景与意义....................................................................................1§1.2基于内容图像检索的系统结构............................................................3§1.3基于...

展开>> 收起<<

基于多维特征向量的图像检索技术的研究.pdf

共55页,预览6页

还剩页未读，继续阅读

基于多维特征向量的图像检索技术的研究

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

推荐作者

热门标签

举报选择: