In recent years, remote sensing image data have increased significantly due to the improvement of remote sensing technique. On the other hand, data acquisition rate will also be accelerated by increasing satellite sensors. Hence, it is a large challenge to make full use of so considerable data by conventional retrieval approach. The lack of semantic based retrieval capability has impeded application of remote sensing data. To address the issue, we propose a framework based on domain-dependent

more »

... tology to perform semantic retrieval in image archives. Firstly, primitive features expressed by color and texture are extracted to gain homogeneous region by means of our unsupervised algorithm. The homogeneous regions are described by high-level concepts depicted and organized by domain specific ontology. Interactive learning technique is employed to associate regions and high-level concepts. These associations are used to perform querying task. Additionally, a reasoning mechanism over ontology integrating an inference engine is discussed. It enables the capability of semantic query in archives by mining the interrelationships among domain concepts and their properties to satisfy users' requirements. In our framework, ontology is used to provide a sharable and reusable concept set as infrastructure for high level extension such as reasoning. Finally, preliminary results are present and future work is also discussed.