【作者】 余芳；

【导师】 郑秋梅；

【作者基本信息】 中国石油大学 ， 计算机应用技术， 2007， 硕士

【摘要】 在基于内容的图像检索中,颜色特征计算简单,具有较好的鲁棒性,作为图像的一种重要视觉信息,已经得到了广泛的应用。本文主要针对图像的颜色空间、颜色的量化和颜色特征提取方法进行了研究。首先,比较了RGB和HSV两种颜色空间模型,对颜色量化方法作了分类介绍。在此基础上重点讨论和比较了常见的颜色特征和颜色空间信息的提取方法,并对颜色特征的相似性度量方法作了介绍。其次,针对颜色直方图缺少颜色的空间分布信息,特征维数较多,存储量大的缺点,作了如下改进:(1)为了获取颜色的空间分布信息,提出一种等面积环形分块方法,该方法不仅保持了颜色直方图的旋转不变性,而且能够突出图像中心的主体地位,符合人的视觉特性。(2)改进了采用K-means算法提取图像主颜色的方法,较好地提取了图像的主颜色直方图,既降低了特征维数,又较好地保持了颜色表示精度。(3)基于上述研究,提出了基于分块主颜色的图像检索方法:对图像进行等面积环形分块,统计各子块的主颜色直方图。该检索方法同时具有等面积环形分块和主颜色直方图的优点。实验证明,分块主颜色方法能明显地改善检索效果。(4)为了进一步提高检索效率,综合分块主颜色方法和分块颜色矩方法进行检索。实验表明,组合检索比单一的检索方法更有效。最后,详细设计和实现基于上述方法的图像检索原型系统IRS,并基于此原型系统,对本文提出的几种图像检索方法进行了实验比较和性能分析。更多还原

【Abstract】 As an important vision information of image, the color features are simple and have better robustness, which has been broadly used in the Content-Based Image Retrieval (CBIR).The expression of image color space, the color quantification and color feature extraction method are analyzed and discussed in this paper.First, it compares RGB and HSV color space model, introduces the color quantification methods. On this basis, the extraction methods of the common color features and the color-spatial distribution information are analyzed and discussed emphatically.Second, the color histogram lacks the color-spatial distribution information, has many feature dimensions and large memory capacity shortcomings. In this paper, the following improvements are made: (1) In order to obtain the color-spatial distribution information, an equal-area ring partition is proposed. The method not only keeps the color histogram rotation invariant, but also focuses on the dominant position of the image center, which is conformed to the visual characteristic. (2) The extraction method of the main color which uses the K-means algorithm is improved. The improved method extracts the main color histogram well, which reduced the number of features and maintained the color expression precision well. (3) Based on the above research, a partition main color image retrieval algorithm is proposed: partition the image using the equal-area ring method and statistic the partition’s main color histogram. This retrieval algorithm has advantages of both the equal-area ring partition and main color histogram. The experiment shows that the partition main color method can improve the retrieval results. (4) To further improve the efficiency, integrated the partition main color method and partition color moment methods. The experiment shows that the combination of search retrieval method is more effective than the single. Finally, an image retrieval prototype system (IRS) is detailed designed and implemented. Based on this prototype system, the image retrieval methods, which are proposed in this paper, are carried on and analyzed the algorithm’s performanceThe theoretical analysis and experimental results show that the partition main color image retrieval method can improve the retrieval accuracy and improve the retrieval results.更多还原