一种融合颜色和纹理特征的遥感图像检索方法

海量遥感图像的自动查询和选择，迫切需要有效的基于内容的图像检索方法。鉴于单一视觉特征不能很好地表达图像内容，为此提出一种基于五叉树分解的线性加权颜色和纹理特征距离的检索新方法。该方法首先采用五叉树分解法分解图像，然后在利用多通道Gabor滤波器与图像做卷积得到滤波能量值的基础上，提取各子图像滤波能量纹理特征，最后通过计算子图像的颜色均值和均方差来对查询图像和与其大小相当的数据库子图像进行颜色和纹理特征线性加权距离相似性测度。将该方法用于高分辨率卫星和航空遥感图像数据库检索的实验结果证明，该方法是有效的。

Remote Sensing Image Retrieval Using Color and Texture Fused Features

Retrieval and management of the vast quantitative image data needs efficient approaches of content-based image retrieval. Current image retrieval methods in geographic image databases use only one kind of image features, which can not describe image content completely. In this paper, a remote sensing image retrieval approach using color and texture fused features is presented. A given image is decomposed using quin-tree, and each subimage except the center one is separated into 5 sublevel subimages until the size of the subimage equals to or is great than 16×16 pixels. The energy values of the image are calculated via multi-channel Gabor filter, and the mean values and standard deviations of each subimage are extracted as texture features. The color features are calculated too. Similarity between a given query image and database subimages, which sizes are approximately equal to the former, are measured using linear weighted distance of color and texture features. Then the top similar subimages are returned as query results. This approach is applied to retrieve high resolution remote sensing images from database, and its efficiency is confirmed by the experiments. The comparison of precision and recall between texture-based and color -texture-based image retrieval shows that synthesized features describe image content better than lonely one does.