基于多特征的图像检索技术研究与应用

【作者】 林丹；

【导师】 刘启和；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2011， 硕士

【摘要】 随着Internet技术的飞速发展,大量图像数据也随之产生,如何快速而准确地从海量的多媒体数据库中检索到用户需要的信息已成为一个亟待解决的问题。基于内容的图像检索技术正是解决该问题的一个发展方向。目前,对该技术的研究已成为一个热门的课题。本文从对图像低层特征的研究入手,研究了图像的颜色特征、纹理特征和形状特征的特性及其提取方法,并建立了一个基于多特征的图像检索系统。在系统建立的基础上,针对多特征检索带来的维数灾难,提出使用PCA、LLE算法对特征向量进行降维,以提高系统性能;在图像的检索过程中引入相关反馈技术,根据用户反馈的信息动态修改特征的权重,以达到实现人机交互,提高系统的检索准确度的目的。本文的主要研究成果有以下几点:1.本文在多特征提取的基础上,实现了一个基于多特征的图像检索系统。并针对不同特征及其组合提供多种查询方法,如基于颜色特征的检索、结合颜色特征和纹理特征的检索;并分析各特征的特点及其在不同类型图像中的表示作用;2.在基于多特征的图像检索条件下,针对高维特征向量带来的维数灾难,将PCA(Principal Component Aanlysis,主成分分析)和LLE(Locally Linear Embedding,局部线性嵌入)技术引入到图像检索的高维向量降维中来,并对二者作出分析比较,在保证查全率和查准率的基础上,实现了有效的降维;3.在基于多特征的图像检索条件下,应用基于特征权重调整的相关反馈技术实现系统与用户的交互,系统根据用户反馈的信息,动态地调整特征向量的权值,提高检索的准确率,使之更能满足用户的需求。更多还原

【Abstract】 With the rapid development of Internet technology, the huge number of image data has been brought. How can users search for the image data that they really need from the vast image database has become a hot problem to be solved. Content-based image retrieval (CBIR) technology is a study that to solve this problem.This thesis has done some researches on low-level image features and their extraction methods, and has achieved a multi-feature-based image retrieval system. To solve the high dimension problem that the Multi-feature-based image retrieval brings, we use PCA or LLE method to reduce the dimensionality. Also, we use relevance feedback technique to know the users’need better and to improve the system’s efficiency. The main work of this thesis is as follows:1. This thesis has carried out a multi-feature-based image retrieval system on the condition of multi-feature has been already extracted. This retrieval system provides many retrieval methods. We can use different kinds of features to retrieval images, such as color feature, or combined with color feature and texture feature. This thesis also has analysed those features and their roles in expressing different kinds of images;2. On the condition of multi-feature-based image retrieval, we put forward PCA and LLE methods to do dimensionality reduction. This thesis also compared this two methods, and improved the system efficiency on keeping the system’s recall and precision;3. On the condition of multi-feature-based image retrieval, we use relevance feedback technique to get the connection of system and users, and adjust the values of the features according to the feedback of users, to raise the accuracy of image retrieval and satisfy the users’need.更多还原