CBIR中特征提取技术的比较研究

发布时间：2018-06-21 19:31

本文选题：图像检索 + 特征提取　；参考：《浙江理工大学》2017年硕士论文

【摘要】：信息化时代,生活中出现了海量的图像信息。要从这些海量信息中检索出与目标相似的图像,一直是图像检索技术研究的目的。以前的图像检索技术主要基于文本。随后,出现了基于图像的颜色、纹理、形状等来提取特征的算法,这些基于语义内容上描述特征的检索方法,即为基于内容的图像检索技术(Content-Based Image Retrieval,CBIR)。基于内容的图像检索中的关键技术之一是特征提取技术。本文首先论述了课题的研究背景和现状,并对基于内容的图像检索技术作了结构性介绍。然后,实验分析了两种传统的特征提取技术在图像检索中的效果。最后,详细介绍了三种发展较好的特征提取方法在图像检索上面的运用,并进行了实验比较分析。本文主要工作如下:(1)在传统的特征提取方法中,本文主要研究了CBIR中基于HSV空间颜色的特征提取方法和基于灰度共生矩阵的特征提取方法。并实验分析了这两种特征提取方法在图像检索方面的检索效果。(2)研究了哈希算法在图像检索系统中的应用。在哈希算法中,主要研究的是均值哈希算法的特征提取技术,并通过离散余弦变换代替图像尺寸缩小对其进行了改进。随后,将改进后的均值哈希算法与改进前的均值哈希算法应用于图像检索中,并对两者的检索效果进行了比较。实验证明,改进后的均值哈希特征提取技术在CBIR中的检索效果要优于未改进的均值特征提取技术和两种传统的特征提取技术。(3)研究了SIFT算法在图像检索系统中的应用。介绍了SIFT算法提取特征描述子的基本原理和具体步骤,并通过实验分析了SIFT特征提取方法在CBIR中的表现。(4)研究了卷积神经网络在图像检索系统中的应用。通过研究经典的卷积神经网络模型,提出一种新型的预训练卷积神经网络来提取图像特征,并通过实验分析了本文卷积神经网络模型在CBIR中的检索效果。为了对各种特征提取方法的性能优劣进行比较,本文中,各种算法都采用相同的实验条件。实验结果表明,与传统的特征提取技术相比,新型的特征提取技术具有更好的检索性能。
[Abstract]:In the information age, mass image information appears in the life. It is the aim of image retrieval technology to retrieve the image similar to the target from these massive information. Previous image retrieval techniques were mainly based on text. Subsequently, there are image based color, texture and shape algorithms to extract features. These semantic content-based feature retrieval methods are Content-Based Image Retrieval (CBIRN). Feature extraction is one of the key techniques in content-based image retrieval. In this paper, the background and present situation of the research are discussed, and the content-based image retrieval technology is introduced. Then, the effects of two traditional feature extraction techniques in image retrieval are analyzed. Finally, the application of three better feature extraction methods in image retrieval is introduced in detail, and the experimental results are compared and analyzed. The main work of this paper is as follows: (1) in the traditional feature extraction methods, this paper mainly studies the CBIR color extraction method based on HSV space and the feature extraction method based on gray level co-occurrence matrix. The effect of these two feature extraction methods in image retrieval is analyzed experimentally. The application of hashing algorithm in image retrieval system is studied. In the hash algorithm, the feature extraction technique of the mean hash algorithm is mainly studied, and the discrete cosine transform is used instead of the image size reduction to improve it. Then, the improved mean hash algorithm and the improved mean hash algorithm are applied to image retrieval, and the retrieval results are compared. Experimental results show that the improved mean hash feature extraction technique is better than the unimproved mean feature extraction technique and two traditional feature extraction techniques in CBIR.) the application of sift algorithm in image retrieval system is studied. This paper introduces the basic principle and concrete steps of feature descriptor extraction of sift algorithm, and analyzes the performance of sift feature extraction method in CBIR through experiments. The application of convolution neural network in image retrieval system is studied. By studying the classical convolution neural network model, a new pre-training convolution neural network is proposed to extract image features, and the retrieval effect of the convolution neural network model in CBIR is analyzed experimentally. In order to compare the performance of various feature extraction methods, the same experimental conditions are used in this paper. Experimental results show that compared with the traditional feature extraction technology, the new feature extraction technology has better retrieval performance.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】