Terrascope image clustering: Applying clustering techniques to image agglomeration in image retrieval systems

In this thesis we describe the design of the Image Clustering features associated with the Terrascope system. TerraScope is an earth science data middleware system that was designed to facilitate collaboration among a set of data repositories (peers) who wish to provide their geospatial data through an integrated portal. Individual Terrascope repositories are designed to locally store images retrieved by satellites from the geographical regions accessible to a particular receiving station. Frequent collection of images from the same geographical region yields images that have a high probability of overlapping with one another making the task of manually browsing through the image database difficult. TerraScope Image Clustering (TIC) attempts to provide an effective way of managing highly overlapping image data by organizing a relational query result set into a tree of image clusters. Each node in the tree represents a cluster or subset of the parent node with less image overlap. Users browse the result set by traversing the cluster tree guided by meaningful labels associated with the clusters. Each cluster in the tree can be recursively organized into finer clusters according to multiple criteria such as collection date, or sensor type. The cluster tree can be changed dynamically and recursively in order to capture the clustering that better suits the user’s information need. The current version of TIC supports several clustering algorithms and was implemented using an