基于内容的词典式结构图像检索

VIP免费
3.0 陈辉 2024-11-19 4 4 2.33MB 55 页 15积分
侵权投诉
摘 要
随着多媒体和网络技术的快速发展,全世界的数字图像的数量正以惊人的速
度增长,图像信息资源的管理和检索显得日益重要。因此直接采用图像视觉内容
CBIR(Content Based Image
Retrieval)成为当前多媒体检索研究的热点之一。CBIR 技术主要根据图像所包
含的颜色、纹理、形状以及对象的空间关系等低层图像特征来分析图像信息,
达图像特征,进行图像间的匹配,进而快速有效地查询和访问相关图像信息。
将多种图像特征综合起来用于图像检索,能够更加全面、准确的表达图像内
容;在此基础上,本文提出了一种综合利用灰度图像矩阵统计值、图像颜色和纹
理特征的图像检索算法—图像词典式结构检索系统,通过提取图像的颜色、纹理
等特征,将待查询图像的特征与图像库中图像的特征进行匹配实现相似图像查
询。
词典式结构是一个分级分类的树形模型。本文构造了这样的一个图像库模
型,首先对图像库中图像进行预处理与分块,然后对子块进行特征向量提取,
成一个特征向量集。这些特征向量全依赖于图像的内容属性,因此,要对图像进
行分类,只需要对这些特征向量进行聚类。本文使用K均值聚类算法,每一级分
类成8个子类,为了尽量使内容上最相似的图像子块聚集在一起,要求构造更多
的子库,但是,如果在一级上构造太多,词典显得庞大,繁琐。为了解决这些问
题,本文提出了一个多级词典结构的图像库的构造,这样图像库中的子块分类空
间成倍数增加,使图像查询可以更好的区分路径,提高了检索效率和安全性。
对于特征向量的选取,先提取由灰度矩阵统计参数、颜色和纹理特性组成的
共11个特征值,另外,用图像块灰度矩阵前两个奇异值之比判断子块图像所含信
息量的大小,并作为特征向量的第12个分量,构成初始特征向量,对于含有较少
的信息量的子块图像将予以舍弃。应用PCA主分量分析法降维,得到代表图像块
的6维主向量。
本文的基于内容的词典式结构图像库的查询,其本质上是对图像分割后的关
键子块进行匹配,然后进行替换和拼接的过程。对于原图像库中存在的图像,
以从词典式图像库中查询匹配到原有图像。对于图像库中不存在的图像,词典库
中将对查询图像每一子块局部匹配,得到与原有图像视觉上极为相似的图像。
典式图像库的子块是按照一定的规则有序地排列在图像库中,图像进行查询时,
每一个子块在图像库中都有一个匹配的路径,词典库中视觉相似图像子块的路径
非常相似,符合图像 Hash 编码的基本条件,我们把图像路径转化为二进制编码,
为词典式结构的图像 Hash 的研究奠定了基础。
最后介绍了非负矩阵分解(NMF)的图像 Hash,并用 NMF 图像 Hash 分析了
查询图像的相似程度,结果表明图像词典式构造的图像库可以准确的匹配出相似
图像。
总之,基于内容的图像词典式结构检索系统具有安全、高效、准确的优点。
关键词:词典式结构 主分量分析 K均值聚类 图像哈希 非负矩阵
分解
ABSTRACT
With the quick development of the multimedia and network technology ,digital
images are increase rapidly with all over the world, it is becoming more and more
important to manage and retrieve images information. Content-based image
retrieval(CBIR) technology which uses image visual content to query image
information becomes most active one in multimedia retrieval fieldCBIR technology
extracts image information through analyzing color texture shape and other
low-layer image feature from imagedescribes image feature informationmatches
images and then enables effectiveefficient query and access Image source.
A multi-features comprehensive retrieval can accurately express the image’s
content; on this basis, the paper presents an image retrieval algorithm with features
comprehensive using gray-scale image matrix statistics, image color and texture
features - Image dictionary structure retrieval system. In order to achieve a similar
image query, we extract the image color, texture and other characteristics to match the
characteristics of the image will be query from dictionary.
Dictionary structure is a tree models with hierarchical classification. This paper
constructs an image database model like this, first, images in a image library are
pre-processed and cut into sub-blocks before using, and then extract features from
blocks, which constitute a feature set based on image content. A large number of
images are classified into 8 sub-categories with k-means clustering. In order to cluster
the images sub-blocks with most similar content together, require more sub-base
structures, but, if constructed more classifications, the dictionary appears large and
complicated. In order to solve these problems, this paper presents a multi-level
dictionary structure, so that the image library space increased Multiply, the distinction
between images to query can find better path which can improve the retrieval
efficiency and safety.
For the feature vector selection, let's extract total of 11 eigenvalues from gray
matrix statistical parameters, color and texture features. In addition, we use the ratio
of the first two singular values of gray-scale image block matrix, the 12th features to
determine the amount of information contained in the image sub-block, the amount of
information that contains fewer sub-blocks the image we have to be discarded.
Application of PCA(Principal component analysis) to reduce dimension to
six-dimensional vector contains main information of image blocks.
Image dictionary structure retrieval system, which is essentially process and
retrieve the key image sub-blocks, and then to replacement and splicing. The original
image exist in library can be match the original image. For the image does not exist in
images dictionary database, the images of each sub-block will match with visually
very similar images to the original image. In addition, sub-blocks in the
dictionary-style image library are depositary with certain rules and orderly arranged in
the image library. when quering, there is a matching path to each sub-block in the
image library, the paths which is visual similar of the images is very similar, In line
with the basic conditions for the image Hash coding, we have the image path into
binary data, laid the foundation for the study of image hash.
Finally, introduce the non-negative matrix factorization (NMF) of the image Hash,
and analysis of the query image similarity, Results show that images of dictionary
structure can accurately match the image library out of similar images.
In short, content-based image dictionary structure retrieval system has avantages of
safe, efficient, and accurate.
Key Words: Dictionary structure Principal component analysis
K-means clusteringImage HashNon-negative matrix factorization.
目 录
摘 要
ABSTRACT
第一章 绪 论............................................................................................................... 1
§1.1 课题的背景及研究意义 ................................................................................... 1
§1.2 基于内容的图像检索的主要实现方式及研究进展 ....................................... 2
§1.3 本课题研究的主要内容 ................................................................................... 4
§1.3.1 基于内容的图像检索系统........................................................................ 4
§1.3.2 图像特征提取与分析................................................................................ 5
§1.4 全文结构 ........................................................................................................... 5
第二章 图像分块与特征提取..................................................................................... 7
§2.1 图像分块 ........................................................................................................... 7
§2.1.1 数字图像的存储形式................................................................................ 7
§2.1.2 图像分块描述............................................................................................ 8
§2.2 图像特征 ......................................................................................................... 10
§2.2.1 灰度矩阵统计特征.................................................................................. 10
§2.2.2 颜色特征.................................................................................................. 12
§2.2.3 纹理特性.................................................................................................. 13
§2.2.4 子块特征向量.......................................................................................... 14
§2.3 本章小结 ......................................................................................................... 15
第三章 PCA 方法词典库特征向量的提取 ...............................................................16
§3.1 PCA 主分量分析涉及的相关数学理论 ......................................................... 16
§3.1.1 协方差和相关系数.................................................................................. 16
§3.1.2 协方差矩阵.............................................................................................. 17
§3.2 PCA 方法原理及算法 ..................................................................................... 17
§3.2.1 PCA 原理及计算公式 ...............................................................................17
§3.2.2 使用 PCA 主分量分析的作用 .................................................................20
§3.2.3 使用 PCA 方法对图像子块特征向量的提取 .........................................20
§3.3 本章小结 ......................................................................................................... 22
第四章 图像主分量聚类分析................................................................................... 23
§4.1 聚类算法 ......................................................................................................... 23
§4.1.1 聚类算法概述.......................................................................................... 23
§4.1.2 聚类技术对图像检索系统效率的优化.................................................. 24
§4.2 K 均值聚类原理 .............................................................................................. 24
§4.2.1 K 均值算法分析 ....................................................................................... 25
§4.2.2 图像检索聚类 K均值算法的实现原理................................................. 25
§4.3 K 均值聚类在图像多特征实例中的应用 ...................................................... 26
第五章 词典式结构图像检索系统的设计............................................................... 28
§5.1 图像词典库结构模型 ..................................................................................... 28
§5.2 图像词典库的构造 ......................................................................................... 29
§5.2.1 图像查询过程.......................................................................................... 30
§5.2.2 图像路径.................................................................................................. 31
§5.3 图像查询结果及分析 ..................................................................................... 32
§5.4 词典式结构在图像 HASH 领域的构想 .......................................................... 36
§5.5 本章总结 ......................................................................................................... 37
第六章 利用 NMF 图像 HASH 对检索结果的验证 ................................................38
§6.1 图像 HASH 技术 .............................................................................................. 38
§6.1.1 图像 Hash 的基本技术要求 .................................................................... 38
§6.1.2 图像 Hash NMF 算法的选取............................................................. 38
§6.2 非负矩阵分解原理 ......................................................................................... 38
§6.2.1 非负矩阵分解描述.................................................................................. 39
§6.2.2 非负矩阵分解的实现.............................................................................. 40
§6.3 利用 NMF 生成图像 HASH ............................................................................ 42
§6.4 汉明距离分析 ................................................................................................. 43
§6.4.1 汉明距离定义.......................................................................................... 43
§6.4.2 实验图像 Hash 的汉明距离分析 ............................................................ 43
§6.5 本章小结 ......................................................................................................... 44
第七章 总结与展望................................................................................................... 45
§7.1 论文的主要工作 ............................................................................................. 45
§7.2 总结与展望 ..................................................................................................... 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......................................... 50
致谢............................................................................................................................. 51
摘要:

摘要随着多媒体和网络技术的快速发展,全世界的数字图像的数量正以惊人的速度增长,图像信息资源的管理和检索显得日益重要。因此直接采用图像视觉内容进行图像信息查询的基于内容的图像检索技术CBIR(ContentBasedImageRetrieval)成为当前多媒体检索研究的热点之一。CBIR技术主要根据图像所包含的颜色、纹理、形状以及对象的空间关系等低层图像特征来分析图像信息,表达图像特征,进行图像间的匹配,进而快速有效地查询和访问相关图像信息。将多种图像特征综合起来用于图像检索,能够更加全面、准确的表达图像内容;在此基础上,本文提出了一种综合利用灰度图像矩阵统计值、图像颜色和纹理特征的图像检索算法...

展开>> 收起<<
基于内容的词典式结构图像检索.pdf

共55页,预览6页

还剩页未读, 继续阅读

作者:陈辉 分类:高等教育资料 价格:15积分 属性:55 页 大小:2.33MB 格式:PDF 时间:2024-11-19

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务
/ 55
客服
关注