Kick-off Meeting on Image Understanding
by Eric Pauwels
To focus the efforts of several ERCIM partners in high-level vision and image processing, a meeting on Image Understanding was organized in February at CWI.
The meeting drew participants from INRIA, KTH (Stockholm), UTIA (Prague), and TCD, as well as endorsements from CNR, SZTAKI, and FORTH. It served as a kick-off meeting for an official ERCIM Working Group, formal approval being expected later this Spring.
Due to a confluence of various technologies, the capturing, storage and manipulation of images or video has become inexpensive and straightforward. As a consequence, vast collections of image-data and video-streams are being amassed in digital libraries rapidly expanding in scope and size. Unfortunately, this evolution might fall victim to its own success as the sheer size and complexity of these digital collections hamper efficient data mining. There is a growing consensus that this new bottleneck can only be alleviated by the development of content-aware processing methodologies that can automatically extract metadata with high semantic value and adapt the processing accordingly. The name of the working group, Image Understanding, has been chosen because it is a broad term that seems to cover all of the above-expounded goals.
The working group intends to organise its activities around three subthemes which by necessity share a lot of common ground. Below we give for each subtheme a non-exhaustive subject list:
- Indexing and retrieval: content-based image and video retrieval, (semi)-automatic generation of semantic metadata, cross-modality data mining (eg, combining images and text)
- Information integration: Fusion of different image modalities, content-aware image enhancement
- Visual decision and control: Visual inspection and expert systems, visual decision and control systems (eg, auto-pilots for cars), intelligent surveillance.
Progress in the above applications will call for methodological advances in research areas such as:
- generic probabilistic models for spatial structure detection and reasoning
- generic mathematical models for spatial processing (eg, morphology, PDEs, wavelets)
- computational intelligence (eg, simulation, Bayesian nets, evolutionary computing).
A scientific workshop will be held as a part of the ERCIM meetings on Thursday 6 June 2002 in Vienna.
Eric Pauwels, CWI
Tel: +31 20 592 4225