Shape similarities measures based on two-level ARG for the retrieval of remote sensing images

Abstract

Content-based image retrieval is widely investigated in image database area. However, work on content-based retrieval for remote sensing images (RSI) is seldom reported in literature. In this paper, a new strategy for content-based remote sensing images retrieval based on shape similarities measures has been proposed, which includes the shape description of each object and spatial relational representation between objects based on the two-level attributed relational graph (ARG). We have also showed that Zernike moments are very effective in describing shape features of each region, and the two-level ARG is efficient for spatial relational information of each disjoint region of an object in RSI. As one of the applications, we have applied this method in our network-oriented multiple satellites images management and online distribution system, and the cloud objects of NOAA AVHRR images are used for verifications. An actual case for the descriptions of a cloud object with Zernike moments, the representations of relations with a two-level ARG and the shape similarity measure results in the database is given.