论文部分内容阅读
对于大型遥感图像数据库,如何快速有效的检索到需要的图像是一个关键问题。虽然许多不同的检索技术被设计用于减少需要检索的目标图像的数量,可是绝大多数检索技术都是基于低级特征并且没有或很少考虑高级语义信息。因此使用这些检索技术,检索到的图像在低级特征空间比较相似而在语义方面却关系不大。为了解决这一问题,本文提出了一种分布式卫星图像检索方案。该方案首先利用贝叶斯网络预选一组与用户查询目的相关的候选图像,然后再利用计算代价更高的基于区域的相似度度量方法来对候选图像重新排序并返回给用户。这样检索到的图像不但与用户的查询目的高度相关,而且与查询图像有着相似的低级信号特征。另外,由于候选图像比数据库中存储的图像要少的多,因此本文提出的检索方案大大减少了对大型数据库的检索时间。
For large remote sensing image database, how to quickly and efficiently retrieve the desired image is a key issue. Although many different retrieval techniques are designed to reduce the number of target images that need to be retrieved, most search techniques are based on low-level features and little or no advanced semantic information is considered. Therefore, using these retrieval techniques, the retrieved images are relatively similar in low-level feature space and semantically. In order to solve this problem, this paper presents a distributed satellite image retrieval scheme. The scheme first uses Bayesian networks to pre-select a set of candidate images related to the user’s query purpose, and then re-sorts the candidate images and returns them to the user by using the more cost-effective region-based similarity measure. The retrieved images are not only highly correlated with the user’s query purpose, but also have similar low-level signal characteristics as the query image. In addition, since the number of candidate images is much less than that stored in the database, the retrieval scheme proposed in this paper greatly reduces the retrieval time for large databases.