Research Of Image Data Management For Space Data Mining

Nowadays, it becomes a hot research topic to manage remote image data effectively after the impressive development of Geographic Information System (GIS) and space data mining technology, especially after the possibility to obtain large scale remote image data. People usually archive remote images into database systems, and manage them with an appropriate index scheme. However, existing index schemes mainly focus on the management of remote images for visualization-oriented applications, while content based index schemes for further space data mining applications are relatively rare in current remote image data management works.This thesis builds a content based index scheme for remote image databases after giving a brief introduction to image databases and the feature extraction of image. This index scheme relies on the image local feature SIFT (Scale-Invariant Feature Transform), then attains SIFT cluster centers called visual words through hierarchical clustering, finally gets a remote image vocabulary tree to index the remote image data. Experiment shows that the proposed index scheme for remote image, which offline constructs a vocabulary tree with thousands of representative remote images, can be completed within one hour.This thesis also takes two space data mining examples, which are the similarity retrieval of one remote image query and the remote image retrieval based on key words, respectively, to demonstrate the effectiveness and efficiency of the proposed content based index scheme for remote image data. It is shown that the remote image database with content based index scheme can respond to data requests of space data mining applications in real time.