Abstract: Remote sensing image is an imagery of surface status within a certain range, which has the properties such as diversity, complexity and mass. That is why the remote sensing image retrieval is often a comparison between target image and sub images of the image database. So the remote sensing image should be decomposed before retrieval. This paper puts forward a new remote sensing image retrieval approach by using improved quintuple tree image decomposition method. First of all, the image is decomposed by improved quintuple tree image decomposition method, which splits large scale remote sensing imagery into sub images. Then texture features of each image block are extracted. Finally, the euclidean distance is computed which is between target image and sub images of the remote sensing image in the image database. These images are returned according to their euclidean distance as the final retrieval results. Aerial remote sensing images during the earthquake of Haiti are used in the experiment, which make a comparison between improved quintuple tree image decomposition method and quintuple tree image decomposition method. The improved quintuple tree image decomposition method is validated using the result of the experiment. It can accurately catch the interested sub images of the user, so a higher recall and precision can be reached.

Key words: content-based image retrieval, remote sensing images, decomposition method of quintuple 2/3 side-length, texture feature extraction

摘要： 由于一幅遥感图像是对一定范围内的地表状态的成像，并且遥感图像具有多样性、复杂性、海量等性质，致使遥感图像检索往往是查询图像和图像库图像的局部区域之间的相似性匹配。为了提高遥感图像的检索效率，必须首先对遥感图像进行分解。提出了一种将遥感图像分层分解的遥感图像检索方法，该方法利用改进五叉树分解法将图像库图像按层次分解成不同大小的子图，在提取子图的纹理特征后，以查询图像和图像库子图之间的欧式距离衡量图像相似度，实现了遥感图像检索。利用海地地震时的航空遥感图像作为实验数据，应用改进五叉树分解法将遥感图像分解后，进行查询检索实验，并与普通五叉树进行了对比。实验结果表明利用改进五叉树分解法进行遥感图像分解后得到的分块图像，可以更精准地查询出用户真正感兴趣的部分，能够获得较高的查全率和查准率，提高查询效率。

关键词: 基于内容的图像检索, 遥感图像, 五叉2/3边长分解法, 纹理特征提取