Image Information System research project

Development activities

The group's major projects:

Progressive Content-based Image Retrieval: This project is jointly funded by NASA/IBM. The main challenges are:

a scalable framework which allows the system to scale to
- large amount of data (accumulated at the rate greater than terabytes/week),
- large amount of features (not much less than the amount of data), and
- large number of users;
an extensible content-based query framework which allows
- user defined features (in addition to texture, spectral histogram, and shape),
- user defined objects (at the semantic, feature or pixel level), and
- composite objects (using spatial, temporal, or boolean relationship).

Our proposed solution includes:

progressive data representation based on a combination wavelet transformation and entropy coding;
progressive feature extraction;
progressive semantic extraction (supervised or unsupervised classification);
progressive search at the semantic, feature and the pixel levels.

Environmental and public health risk assessment ,
Precision agriculture,
Medical diagnostic assistance,
Petroleum core images and seismic data analysis.

Please take a look at our demo system and the white paper.

Interactive Query Construction: This project tries to address three emerging issues faced by the multimedia applications with unstructured data (such as time series, images, and video) in a client/server environment:

Interactive query construction: the necessary information regarding to the valid query syntax schema, attributes, valid data range is available to the user right before the user needs it,
Query parsing offloading: the client (usually written in Java) performs as much as possible the checking of the syntax and semantics (YES! semantics) of the query.
Extensible query environment: where the user can incorporate user defined features, user defined objects and composite objects using the same query framework (just like SQL).
Iterative refinement: the user can easily incoporate both previous queries and query results into a new query.