基于Curvelet特征应用的太阳图像检索研究

发布时间：2018-04-18 18:55

本文选题：CBIR + Curvelet变换　；参考：《昆明理工大学》2017年硕士论文

【摘要】：近年来,挖掘大型图像库(比如太阳图像)所包括的内容依然是很具有挑战性的,而且图像库也在一直不断的增加。因此,想要在相关图像库中找到特定的那些图像变得越来越困难。想要获取具体的特定图像,而这些图像却被埋葬在了庞大的图像库中。例如,在成千上万张图像的太阳图像库当中,想要获取特定的某种类型的照片依然不是个简单的任务。检索系统中所使用的搜索引擎必须既高效又有效,而基于内容的图像检索技术(CBIR)刚好满足这样的要求,它是利用图像的颜色、纹理、形状、边沿等底层视觉特征来描述图像内容并进行检索的,CBIR克服了基于文本的图像检索方法的诸多缺点,引起研究人员的广泛关注。而对于太阳图像这种灰度图像来说,选择纹理特征来描述无疑是效果最好的。在以往的图像处理过程当中小波变换因其具有较好的时域频域分析能力,而被人们广泛使用,但是小波变换也不是处处都是有优势的,当进行超平面奇异性的处理时,小波变换就显得不是特别合适了。而曲波(Curvelet)作为一种新的分析工具,它能够稀疏的表示具有直线特征的图像,可以应用到二维图像处理过程中。本文首先在研究多尺度几何分析以及Curvelet理论的基础上,对Curvelet变换的实现过程进行了研究与分析。在此基础上,论文主要做了以下工作:第一,从基于内容的图像检索入手,通过对CBIR系统结构的分析,针对本文所采用的太阳图像样本库,选择纹理特征提取方式,查询过程中匹配的时候选择欧氏距离相似性度量方式,以及对结果评价的时候选择查全查准曲线,来进行检索实验。第二,将Curvelet纹理特征提取之后应用在太阳图像的检索过程中,并且通过检索实验来验证其检索的性能,接着又对Curvelet特征提取算法提出了优化,成功的对Curvelet特征进行降维,提高了检索效率。第三,将Curvelet纹理特征与Wavelet纹理特征进行简单加权融合,将融合之后的特征进行检索实验,融合之后的特征带来了不错的检索性能。
[Abstract]:In recent years, mining large image databases (such as solar images) is still very challenging, and the image library has been increasing.Therefore, it becomes more and more difficult to find specific images in the relevant image library.Want to obtain specific images, and these images are buried in a large image library.For example, in the solar image library of thousands of images, it is still not an easy task to get a certain type of photo.The search engine used in the retrieval system must be both efficient and effective, and the content-based image retrieval technology (CBIR) just meets this requirement. It uses the color, texture, shape of the image.Edge and edge visual features to describe the image content and retrieve the CBIR overcomes many shortcomings of text-based image retrieval method and attracts wide attention of researchers.For the grayscale image of solar image, it is the best to choose texture feature to describe it.In the previous image processing process, wavelet transform is widely used because of its good analysis ability in time domain and frequency domain, but wavelet transform is not always advantageous, when processing hyperplane singularity,Wavelet transform is not very suitable.Qu Bo Curvelet, as a new analysis tool, can sparse represent images with linear features, and can be applied to two-dimensional image processing.In this paper, based on the study of multi-scale geometric analysis and Curvelet theory, the realization process of Curvelet transform is studied and analyzed.On this basis, the thesis mainly does the following work: first, starting from the content-based image retrieval, through the analysis of the CBIR system structure, aiming at the solar image sample database adopted in this paper, we choose the texture feature extraction method.The Euclidean distance similarity measurement method is chosen to match in the query process and the full search curve is selected to carry on the retrieval experiment when the result is evaluated.Secondly, the Curvelet texture feature extraction is applied to the retrieval process of the solar image, and the retrieval performance is verified by the retrieval experiment. Then the Curvelet feature extraction algorithm is optimized to reduce the dimension of the Curvelet feature successfully.The retrieval efficiency is improved.Thirdly, the Curvelet texture feature and the Wavelet texture feature are simply weighted fusion, and the fusion features are used in the retrieval experiment. The fusion features bring good retrieval performance.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】