基于尺度不变特征的图像检索技术研究

发布时间：2018-04-18 12:08

本文选题：图像检索 + 尺度不变特征　；参考：《西安科技大学》2017年硕士论文

【摘要】：随着数码科技以及互联网通信的迅速发展,数字图像已经渗透到人们的日常生活中,如交通监控、医疗卫生以及新闻媒体等。而大多数数字图像都以无序的状态存在,图像中的一些重要信息得不到利用,寻找一种高效的检索算法让用户可以准确、快速地获取所需目标图像,已成为一项急需解决的问题。针对上述需求,本文首先对目前所使用的检索算法进行阐述,分析传统的图像检索方式并着重对基于内容的图像检索技术进行研究。传统的图像底层特征在光照、尺度、旋转等情况中,无法保障检索结果的稳定性,因此本文选用经典的局部尺度不变SIFT算法作为图像检索的算法。研究发现,上述算法存在误匹配点较多、无法在实际中应用等缺点。因此本文对此算法进行改进,一是为图像SIFT特征描述符构建KD-tree数据索引,并使用最优节点优先搜索BBF算法提高图像特征点对的配准速率;二是结合改进的最近邻匹配算法与随机抽样一致算法进一步去除错误匹配点对。最后,在多种实验环境中验证改进的SIFT算法,结果表明改进的SIFT算法在保留原始算法优点的同时,可以有效的降低误匹配率,能在基于内容的检索系统中得到更好的应用。基于改进的SIFT算法,本文又设计并实现了一个简易基于内容的图像检索系统,该系统主要由数据库建立和数据库查询两部分构成。通过对比分析表明,改进的SIFT算法应用在实际检索系统中不仅能有效减少配准时间,也可以提高图像的配准率。
[Abstract]:With the rapid development of digital technology and Internet communication, digital images have penetrated into people's daily life, such as traffic monitoring, health care and news media.However, most digital images exist in a disordered state, and some important information in the image is not utilized, so a high efficient retrieval algorithm is found to enable users to obtain the desired target image accurately and quickly.It has become an urgent problem.In order to meet the above requirements, this paper first describes the retrieval algorithms used at present, analyzes the traditional image retrieval methods and focuses on the research of content-based image retrieval technology.The traditional image bottom features can not guarantee the stability of retrieval results in illumination scale rotation and so on. Therefore the classical local scale invariant SIFT algorithm is chosen as the image retrieval algorithm in this paper.It is found that the algorithm has many mismatch points and cannot be applied in practice.Therefore, this algorithm is improved in this paper. One is to build the KD-tree data index for the image SIFT feature descriptor, and to use the optimal node-first search BBF algorithm to improve the registration rate of the image feature points.Secondly, the improved nearest neighbor matching algorithm and the random sampling matching algorithm are combined to remove the error matching points.Finally, the improved SIFT algorithm is verified in a variety of experimental environments. The results show that the improved SIFT algorithm can effectively reduce the mismatch rate while preserving the advantages of the original algorithm, and can be better applied in content-based retrieval system.Based on the improved SIFT algorithm, a simple content-based image retrieval system is designed and implemented in this paper. The system consists of two parts: database establishment and database query.The comparison and analysis show that the improved SIFT algorithm can not only reduce the registration time but also improve the image registration rate in the actual retrieval system.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】