一种改进包模型的遥感图像检索方法  

An Improved Remote Sensing Image Retrieval Method Based on Bag of Word Framework

作　　者：杨进[1] 刘建波[1] 戴芹[1] 

机构地区：[1]中国科学院遥感与数字地球研究所,北京100094

出　　处：《武汉大学学报（信息科学版）》2014年第9期1109-1113,共5页Geomatics and Information Science of Wuhan University

摘　　要：提出了一种改进的包模型图像检索方法,使用本底图像构建视觉词汇字典。在建库过程中,通过近似KD-Tree搜索本底图像局部不变特征与入库图像特征间的对应匹配关系,并记录该关系;图像检索时,搜索检索图像与本底图像特征的对应关系,进而计算与入库图像的关系。该方法仅保存本底图像的局部不变特征,可实现感兴趣区域的图像检索。实验结果表明,该方法在提高图像检索准确性的同时,可以减少所需存储局部不变特征的数量。The Bag-of-Words （BOW） approach has lower retrieval accuracy performance for remote sensing image retrieval. The rate of correct results is very low especially when target textures are similar. To overcome this shortcoming, an indirect comparison between a query and target image based on local invariant feature of the background image is proposed. The solution has two phases： the process of building a database and a retrieval process. After building a database, a search of the nea rest neighbor feature in a query image in the feature space of the image dataset through an approximate KD-Tree is executed, then their relationship is recorded . The relationship can be used in calculating similarity between a query and target image during a re,rieval procedure. Experiments show that the proposed method has a better recognition performance than that of the BoW and, at the same time, needs less storage space for saving local invariant features.

关 键 词：遥感图像检索 本底图像 包模式 视觉词汇 感兴趣区 

分 类 号：P237.4[天文地球—摄影测量与遥感] TP753[天文地球—测绘科学与技术]