SIA OpenIR  > 光电信息技术研究室
基于内容的图像检索中视觉描述方法的研究
Alternative TitleResearch on Visual Description Methods in Content-based Image Retrieval
吴永国1,2
Department光电信息技术研究室
Thesis Advisor罗海波
ClassificationTP391.4
Keyword图像检索 特征提取 全局描述 局部描述
Call NumberTP391.4/W86/2011
Pages64页
Degree Discipline模式识别与智能系统
Degree Name硕士
2011-05-27
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract基于内容的图像检索是一种依靠图像的颜色、纹理和形状等视觉内容来实现图像自动检索的技术。视觉内容的描述是其最为关键的环节,直接影响系统的检索效果。本文着重研究了视觉内容的描述方法:全局描述、局部描述以及融合全局和局部的描述。 第一部分为全局描述方法。详细介绍了MPEG-7主颜色描述算子和边缘直方图描述算子,针对单一特征不能全面地描述图像的内容的问题,提出了基于MPEG-7的主颜色和边缘直方图的多特征描述方法,实验结果表明,基于多特征的描述方法要优于单一特征的描述方法。 第二部分为局部描述方法。为克服全局特征缺乏语义性的缺点,提出一种基于K-means的自适应图像分割算法,将图像分割成几个不同区域,分割后的区域对应于物体或物体的一部分,具有一定的语义性。在局部描述匹配时,根据图像检索任务的要求,提出了一种改进的Hausdroff距离。实验结果表明,局部描述方法优于全局描述方法。 第三部分为融合全局与局部描述的图像描述方法。基于前面两部分的研究,发现全局描述缺乏特征的空间位置信息,对图像的细节描述能力很差;基于分割的局部描述中,分割是关键的环节,目前尚未有一种符合人理解的图像分割算法,分割问题阻碍了图像检索的效果,且图像分割算法一般耗时较长,难以实时实现。为此,提出了一种融合全局和局部描述的图像检索算法,该算法的思想是通过融合全局描述和局部描述实现从不同的细节来描述图像内容。其局部区域通过固定分块获得,省去了分割步骤,提高了计算速度。经过测试,本算法取得了较高的检索率。
Other AbstractContent-based image retrieval (CBIR) is one kind of automatic retrieval technology based on visual contents such as color, texture and shape. Visual content description is the most critical aspect of CBIR and direct impact on the system's retrieval results. This paper focuses on the visual content description methods: global description, local description and the fusion of global and local description. The first part is global description. Introducing the MPEG-7 dominate color descriptor and edge histogram descriptor in detail. With respect to the problem that a single feature could not represent content of image completely. Proposed a multi-feature description method based on dominate color descriptor and edge histogram descriptor. Experiments result show that the multi-feature description is much better than the single feature description. The second part is local description. With respect to the problem that global feature lack of semantic, proposed an adaptive image segmentation algorithm based on K-means. An image could be segmented into several regions which correspond to the object or parts of object. Local feature is extract from the regions, so the local description has more semantic than global description. According to image retrieval tasks, an improved Hausdroff distance metric is proposed to compute the similarity using the local description. The experiments results demonstrate that local description has higher retrieval rate than global description.  The third part focused on method combining global and local description. Based on previous work, we find that global description lack of the information about the spatial feature distribution and the power to descriptive details is very poor. In the local description, image segmentation is an essential procedure, and at presents none of image segmentation algorithm accords with human understanding. The segmentation problem hindered development of image understanding. And image segmentation algorithms are usually time-consuming and difficult to real-time implementation. Therefore, a visual description method combining global description and local description is proposed. The basic idea of this algorithm is described image content from the different details. The local blocks are obtained through the fixed partitioned that dispensing with segmentation steps and improves the calculation speed. The result of experiments indicates that this description method can achieve high retrieval rate.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/9237
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
吴永国. 基于内容的图像检索中视觉描述方法的研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2011.
Files in This Item:
File Name/Size DocType Version Access License
基于内容的图像检索中视觉描述方法的研究.(12711KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[吴永国]'s Articles
Baidu academic
Similar articles in Baidu academic
[吴永国]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[吴永国]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.