Developing a Scientific Image Retrieval System with Prediction Capability

Observing scientific images allows scientists to have better understanding on Earth. For example, remote sensing images are applied to help monitoring ocean and climate change. These images will be stored in database and may be retrieved later for reference purpose. The possible retrieval methods may be to give keywords or a query image to the system to get desired images. However, scientists sometimes might want the scientific image system be smart, that is, they want the system to present some images that can predict future variations. Currently, the retrieval methods can not achieve the previous requirement. This paper proposes an intelligence scientific image retrieval system based on content-based image retrieval. While retrieving, this system can exploit the spatial relationship between the objects in query image, and then retrieve images that their spatial semantic are similar to the query image. Moreover, the spatial association rules are also mined and are then used as a basis for retrieving additional images for prediction purpose. The experiments demonstrate that the images retrieved by the proposed system are more accurate, and the system can also predict the future variations.