【作者】 徐艳；

【导师】 王加俊；

【作者基本信息】 苏州大学 ， 通信与信息系统， 2005， 硕士

【摘要】 基于内容的图像检索是多媒体信息处理中的一个重要问题。图像、视频作为多媒体中最直观、最形象的内容,对它们的检索和查询是多媒体信息处理的一个重要方面。本文首先回顾了基于内容的图像检索理论及其应用研究的整个发展过程,全面综述了在图像检索,尤其是基于内容的图像检索领域的技术和现状,并探讨了图像检索中的一些关键技术。纹理是图像最重要的特征之一,在很多基于内容的图像检索中都作为重要的特征用来匹配与查询。本文首先使用基于共生矩阵的纹理特征进行检索,为了提高检索准确率,提出综合纹理和像素中心的检索方法。对花卉、植物的图像库的实验表明,该方法比使用单一纹理特征检索的效果要好。压缩域的图像检索是基于内容的图像检索的一个热点,本文在这方面也作了一定的探讨和研究,提出了使用拼贴误差直方图的分形图像检索方法。拼贴误差是值域块和“最匹配的”定义域块相似性的一个度量,且直方图使用统计特征,具有计算简单的特点。所以该方法不仅有一定的检索准确性,最重要的是大大减少了计算复杂度。实验结果证明,该方法的确既能使计算简便,又能保证检索的准确性。更多还原

【Abstract】 Content Based Image Retrieval (CBIR) is an important issue in multimedia information processing. As the most direct and vivid content, retrieval of image and video information is an important aspect in multimedia information processing. In this thesis, firstly a brief review of the theory, applications and development of CBIR are presented, after which, the techniques and the current state of image retrieval, especially the state of CBIR are summarized. Finally, some important techniques in Content Based Image Retrieval System (CBIRS) are introduced. Texture is one of the most important features of an image. Almost all of the presented CBIR systems use texture as an essential feature for matching and retrieval. In this thesis, a method utilizing both texture features and image center is proposed for retrieval. Experimental results on image database of real flowers and plants show that this method out-performs the method only with texture features based on the co-occurrence matrix. At the same time, CBIR techniques in the compressed domain have attracted much interest. A new fractal coding based indexing technique with histogram of collage errors as the retrieval keys is proposed. Collage error is a quantitative measure of the similarity between the range block and the “best-matched”domain block. Meanwhile, histograms can capture statistical characteristics and can be easily computed. So the proposed method can not only reduce computational complexities greatly, but also enhance the retrieval accuracy. Experimental results on a database of more than 200 texture images illustrate that the proposed method performs excellently.更多还原