局部特征和全局特征相融合的图像检索算法研究

发布时间：2018-03-03 02:09

本文选题：图像检索　切入点：局部特征　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着互联网+时代的来到,网络与人们的学习、生活以及工作等都发生了越来越紧密的联系,与此同时也使得图像数量急剧增加。那么,如何从海量的图像数据库中极速、精确地检索图像显然已成为一个十分有意义而且具备挑战性的课题。目前,图像检索技术主要分为两大类:基于文本的图像检索(TBIR)和基于内容的图像检索(CBIR)。图像检索研究谈论的核心技术是图像特征的提取。图像的特征一般可以通过局部特征或者全局特征来描述,全局特征可以把图像各个部分的影响因素综合的考虑进去,而局部特征更能反映出细节信息。然而,图像内容的变化是如此复杂和多变,要想仅通过单一特征就能够高效准确地完成图像检索恐怕是非常困难的。针对这一问题,本文提出一种新的基于局部特征和全局特征相融合的图像检索算法。主要研究工作如下:1.针对图像局部特征和全局特征如何提取进行研究和讨论。对于局部特征,选取最具代表性的SIFT特征,并着重介绍一幅图像特征描述子的生成过程;对于全局特征,分别讨论量化以后的颜色特征和改进的LBP纹理特征。2.针对SIFT描述子大多是使用图像的灰度信息来进行描述,对形状类似但颜色不同、同一形状但背景有差异的物体辨别能力差的问题,本文提出一种改进的SIFT特征描述符,并在改进的基础之上融合颜色特征对图像进行检索。3.针对SIFT算法在图像特征提取和检索中易受图像噪声和亮度变化影响的问题,本文提出了基于纹理的LBP算法与局部SIFT算法相结合的图像检索算法。通过大量实验仿真表明,本文提出的融合算法与仅靠单一特征检索图像的算法比较,具有很好的检索性能,证明了基于局部特征与全局特征的图像检索算法的准确性和有效性。
[Abstract]:With the advent of the Internet era, the network and people's study, life and work have been more and more closely linked, and at the same time, the number of images has increased dramatically. Accurately retrieving images has clearly become a very meaningful and challenging subject. Image retrieval technology is divided into two main categories: text-based image retrieval (TBIRR) and content-based image retrieval (CBIR). The core technology of image retrieval research is image feature extraction. Feature or global feature to describe, The global feature can take into account the influence factors of each part of the image, while the local feature can reflect the detail information more. However, the change of the image content is so complex and changeable. It may be very difficult to do image retrieval efficiently and accurately with a single feature. In this paper, a new image retrieval algorithm based on the fusion of local and global features is proposed. The main research work is as follows: 1. How to extract local and global features of images is studied and discussed. The most representative SIFT feature is selected, and the process of generating a feature descriptor of an image is introduced. The color features after quantization and the improved LBP texture features are discussed respectively. For the SIFT descriptors, most of them are described by the gray level information of the image, but the shapes are similar but the colors are different. In this paper, an improved SIFT feature descriptor is proposed for the problem of poor discriminant ability of objects with the same shape but different background. Based on the improved algorithm, the color feature is fused to retrieve the image. 3. Aiming at the problem that SIFT algorithm is easy to be affected by image noise and brightness change in image feature extraction and retrieval. In this paper, an image retrieval algorithm based on texture based LBP algorithm and local SIFT algorithm is proposed. A large number of experiments show that the fusion algorithm proposed in this paper has good retrieval performance compared with the one based on a single feature. The accuracy and validity of image retrieval algorithm based on local feature and global feature are proved.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】