基于显著性区域和基元共生矩阵特征的图像检索研究

发布时间：2018-07-26 19:10

【摘要】：随着图像数据呈现几何级数的快速增长,如何实现对图像数据库更加高效、准确的检索,是众多学者研究的目标和方向。基于内容的图像检索通过提取图像的颜色、纹理、形状等底层特征进行相似度匹配,快速地查找目标图像,该技术是当前信息检索领域中的一种重要手段。传统的基于内容的图像检索技术存在一定的局限性,一方面,为了提高对图像主体内容的高度概括性信息描述,盲目的过滤掉图像当中的冗余信息,忽略了不相关部分信息的重要性;另一方面,图像往往包含了丰富的内容信息,使用单一的特征描述只能表达图像的某一方面信息,无法诠释图像的完整性信息描述。颜色特征作为数字图像的一种重要特征描述,因其计算简单有效,被广泛的应用于图像特征的提取和索引。现有的基于内容的图像检索技术通过融入多个底层特征进行图像检索,从而减少了对原始图像描述的信息缺失并提高了图像检索的准确率,但同时也增加了算法的计算复杂度。为了更好的描述图像的内容,本文通过将颜色特征作为图像的全局描述,然后利用有效地特征加权融合方式,分别结合了显著性区域颜色特征和基元共生矩阵特征。通过从不同侧面对图像的内容进行特征描述,达到了较好的检索效果。本文主要针对显著性区域特征和基元共生矩阵特征进行了相关的改进,并在图像分割、特征提取、特征融合等方面进行了部分优化。主要内容如下:1.本文首先深入研究了基于内容的图像检索领域中的相关技术,并针对图像的颜色、纹理、形状等特征的提取方法进行了分析对比。然后详细介绍了基于颜色特征的图像检索,和一些常用的相似性度量方法、性能评价指标。最后针对显著性区域和基元共生矩阵特征的局限性问题,提出了两种有效的图像检索方法,分别为基于全局和显著性区域颜色特征的图像检索方法和基于分块基元共生矩阵的图像检索方法。2.针对显著性区域特征忽略背景信息的问题,提出了一种基于全局和显著性区域颜色特征的图像检索方法(GASCH),在不丢失背景信息的前提下突出强调了显著性区域特征的重要性。首先,提取量化的HSV颜色直方图特征作为全局描述。其次,利用一种显著性区域检测方法将显著性区域和背景区域分离。之后,提取显著性区域的颜色直方图用于构成局部描述。最后,本文通过使用一种自适应加权的方法结合了这两种描述。在Corel 1000图像库中的实验表明,该方法与单一特征检索方法相比具有更好的视觉效果,检索准确率也至少提高了9%。3.针对基元共生矩阵中不满足平移不变性和不同子块可以用相同基元表示的局限性问题,提出了一种基于分块基元共生矩阵的图像检索方法(BMCM)。该方法,首先将图像进行区域划分,并对各个区域分别提取量化的HSV颜色直方图特征,基元共生矩阵特征以及局部二值模式特征。考虑到不同特征描述了图像的不同属性和内容,本文通过对各个特征检索的相似度进行加权融合,实现了颜色、纹理以及空间信息等多特征的图像检索。在Corel 1000标准图像库的实验结果表明,与MCM、BCTF和MCMCM算法相比,该方法有较高的检索准确率且在部分图像类别的检索中有比较明显的优势。
[Abstract]:With the rapid growth of the geometric progression of the image data, how to achieve more efficient and accurate retrieval of the image database is the goal and direction of many scholars. The traditional content based image retrieval technology has some limitations. On the one hand, in order to improve the highly generality information description of the subject content of the image, the redundant information in the image is filtered blindly, and the importance of the unrelated information is ignored; on the other hand, the image goes to the image. It contains a rich content information, using a single feature description only to express a certain aspect of the image information, can not interpret the image integrity information description. Color features as an important feature of the digital image description, because of its simple and effective calculation, it is widely used in the extraction and index of image features. In order to better describe the content of the image, the color feature is used as the global description of the image, in order to better describe the content of the image, the image retrieval technology can reduce the missing information of the original image description and improve the accuracy of the image retrieval. Then, using the effective feature weighted fusion method, the color feature of the significant region and the characteristics of the basic element symbiosis matrix are combined respectively. By describing the features of the images from different sides, the better retrieval results are achieved. This paper mainly improves the characteristics of the significant regional characteristics and the basic element symbiotic matrix. The main contents are as follows: image segmentation, feature extraction, feature fusion, and so on. The main contents are as follows: 1. firstly, the related technologies in the field of content based image retrieval are studied in this paper, and the extraction methods of color, texture and shape are analyzed and compared. Then the color features are introduced in detail. Image retrieval, and some commonly used similarity measurement methods, performance evaluation indexes. Finally, two effective image retrieval methods are proposed, which are image retrieval methods based on global and significant regional color features and graph based on the partitioned primitive symbiotic matrix, aiming at the limitation of the characteristics of significant region and basic element symbiotic matrix. The image retrieval method (.2.) has proposed a image retrieval method (GASCH) based on global and significant regional color features, which emphasizes the importance of significant regional features without losing background information. First, extracting quantized HSV color histogram features as a whole. Second, a significant regional detection method is used to separate the significant region and the background region. Then, the color histogram of the extracted significant region is used to form a local description. Finally, this paper combines the two descriptions by using an adaptive weighting method. The experiment in the Corel 1000 image library shows that this method is used. Compared with the single feature retrieval method, it has better visual effect, and the retrieval accuracy also improves at least the limitation of 9%.3. against the translation invariance in the basic element symbiosis matrix and the representation of different sub blocks with the same basic elements. A image retrieval method based on the partitioned primitive symbiotic matrix (BMCM) is proposed. First, the image is divided into regions, and the quantized HSV color histogram features, the feature of the base element symbiotic matrix and the feature of the local two value pattern are extracted respectively. Considering the different features and the contents of the images, the similarity degree of each feature retrieval is weighted to realize the color and texture. The experimental results of the Corel 1000 standard image library show that compared with the MCM, BCTF and MCMCM algorithms, the method has a higher retrieval accuracy and more obvious advantages in the retrieval of some image categories.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】