W/NV GRSS Chapter Technical Meeting
at George Mason University (Research 1, Room 301)
on Thursday, September 28, 2006 at 4:30 p.m.
Professor Mihai Datcu, IEEE Senior Member
German Aerospace Center, DLR, Oberpfaffenhofen, Germany
Dr. Klaus Seidel
Swiss Federal Institute of Technology, ETH, Zurich, Switzerland
Image Information Mining: the semantic gap
The widespread availability of high resolution EO imagery gives rise to volumes of data but also brings orders of magnitude of image detail and enormously increased information content. Heterogeneous data supporting the interpretation of EO imagery, e.g., multimedia, scientific and engineering measurements, is also continuously generated and stored. However, communicating the information content of such data to people for use in practical applications is still limited by current data processing concepts and technologies. In this contribution, we present novel methods for making inferences using EO imagery by objectively and systematically identifying specified characteristics of images and implementing these algorithms in a new concept for knowledge-driven image information mining and scene understanding. The concept enables the communication of information from a very large image repository of data to users. The communication is at a semantic level of representation and is adapted to the userís conjecture by storing data in form of text that has implicit meaning (i.e., semantic significance).
Specifically, we present novel theoretical concepts and collaborative methods for:
†† *†† Extraction and exploration of the content of
large volumes of high
†† *†† Establishing the link between the user needs
and knowledge and the
information content of images
†† *†† Communicating at a high semantic abstraction between heterogeneous†††† sources of information and users with a very broad range of interests
†† *†† Accessing intelligently and effectively the information content in large EO data repositories
†† *†† Improved exploration and understanding of Earth structures and processes
†† *†† Increasing the accessibility and utility of EO data.
The presentation provides a new perspective on methods for information extraction, sensor and information fusion, machine learning, understanding of user conjecture, and related supporting technologies, e.g. semantic image indexing, categories and ontology generation, etc. The presentation will also cover examples and online demos using a broad variety of data, including high resolution synthetic aperture radar (SAR) and meter resolution optical and hyperspectral imagery. Several examples will address the class of medium-resolution optical images.