基于多特征融合与SVM的图像检索研究

发布时间：2019-01-03 08:52

【摘要】：随着互联网技术与多媒体技术的飞速发展以及大数据时代的到来,网络上的图像数据呈爆炸式的增长趋势,人们的日常生活中也随之充斥着各种各样的图像信息。如何从海量的图像数据中快速而且准确的获取图像信息成为一个迫切的需求。基于内容的图像检索技术正是解决这一问题的研究重点与方向。然而基于内容的图像检索的性能又受着图像底层特征与图像上层理解之间的语义鸿沟的影响。论文从两个方面对图像检索技术开展研究:第一部分研究基于多特征融合的图像检索;第二部分研究基于SVM(支持向量机)相关反馈的图像检索。主要工作如下:(1)提出了一种多特征融合的图像检索算法。首先从图像的底层特征提取方法展开研究,从整体上描述一幅图像,提取全局图像特征GIST,然后采用相似性度量方法在数据库中寻找与查询图像在场景上相似的匹配图像。之后以图像局部特征为出发点,提取场景相似匹配图像的局部SIFT(尺度不变特征变换)特征,由于一副图像的局部SIFT特征为若干个特征点,因此采用基于BBF(Best Bin First)搜索算法的点对匹配算法进行SIFT特征点匹配。最后根据查询图像与场景相似匹配图像SIFT特征点匹配数量返回检索结果。通过先整体后局部的思想融合GIST特征与SIFT特征实现图像检索,本文融合算法与单特征算法的检索性能进行了实验对比,实验结果表明,融合多特征的检索算法性能要优于基于单特征的检索算法。(2)在传统的基于内容的图像检索基础上研究了基于SVM相关反馈的图像检索。将机器学习的思想加入到图像检索中,在讨论相关反馈技术及支持向量机技术后,采用传统的方法进行图像检索,然后根据SVM相关反馈技术再进行检索,从而得到符合用户检索意图的结果。实验表明,基于SVM相关反馈的图像检索取得了有效的检索结果,在图像检索中具有一定的实用价值。
[Abstract]:With the rapid development of Internet technology and multimedia technology and the arrival of big data era, the image data on the network is explosive growth trend, people's daily life is also filled with a variety of image information. How to obtain image information quickly and accurately from massive image data becomes an urgent need. Content-Based Image Retrieval (CBIR) is the research focus and direction to solve this problem. However, the performance of content-based image retrieval is affected by the semantic gap between the underlying features of the image and the understanding of the image upper layer. In this paper, image retrieval technology is studied from two aspects: the first part is image retrieval based on multi-feature fusion, and the second part is image retrieval based on SVM (support vector machine) correlation feedback. The main work is as follows: (1) A multi-feature fusion image retrieval algorithm is proposed. Firstly, the method of feature extraction from the bottom layer of the image is studied. An image is described as a whole, and the feature GIST, of the global image is extracted. Then, the similarity measure method is used to find the matching image similar to the query image in the scene in the database. Then the local SIFT (scale invariant feature transformation) feature of the scene similar matching image is extracted from the local feature of the image. The local SIFT feature of a set of images is several feature points. So the point pair matching algorithm based on BBF (Best Bin First) search algorithm is used for SIFT feature point matching. Finally, the retrieval results are returned according to the number of SIFT feature points of the query image and scene similar matching image. Image retrieval is realized by fusion of GIST features and SIFT features. The performance of the fusion algorithm is compared with that of the single feature algorithm. The experimental results show that, The performance of multi-feature based retrieval algorithm is better than that of single feature based retrieval algorithm. (2) based on traditional content-based image retrieval, image retrieval based on SVM correlation feedback is studied. The idea of machine learning is added to image retrieval. After discussing the technology of correlation feedback and support vector machine, the traditional method is used for image retrieval, and then the retrieval is carried out according to SVM correlation feedback technology. The result is in accord with the user's search intention. The experimental results show that the image retrieval based on SVM correlation feedback has achieved effective retrieval results and has certain practical value in image retrieval.
【学位授予单位】：中南林业科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】