KES Project

Project Title   Knowledge Enabled Services
Project Acronym   KES
Contractor(s)   Advanced Computer Systems Spa, DLR


Project Objectives
          How it Works


Objectives   Top

Identify the technologies, and demonstrate them through a prototype at different deepness levels, applicable to a number of fields required to support image information mining and related user interactions:

Image   learning systems
Image   knowledge acquisition and sharing within user communities
Image   image interpretation support
Image   multi-type data handling
Image   semantic interactions
Image   automatic system adaptation to user behaviour

The use of COTS will be complemented by specific software development to fulfil all the complex features requested to the system.



Context   Top

KES should permit the acquisition of knowledge from users with different levels of expertise and the retrieval of multi-type data on the basis of single queries.

Compared with other data types (e.g.: structured data) the field of Image and spatial data analysis and understanding reaches much higher level of complexity because of:

Image   the huge volume of data (GB to TB)
Image   the variability and heterogeneity of the image data (diversity of sensors, time or conditions of acquisition, etc.)
Image   the image content (its meaning), which is many times subjective, depending on user interest
Image   the large range of user interests, semantics and contextual (semiotic) understanding

Some factors seem to determine the low use of EO images, particularly in other communities (universities, research centres, civil protections):

Image   the tools used by image analysts and experts differ across user communities
Image   these tools are in some cases expensive and not easy to use
Image   large experience (knowledge) is required to correctly interpret images and detect embedded features
Image   image interpretation needs information provided by GIS and other data bases when the volume of data becomes large

Tools like KES could offer a new way to access and interpret the images, not just permitting to select and show the images, but also providing an e-collaboration environment for:

Image   selection and provision of multi-type data (text, GIS layers, images)
Image   shared image interpretation and value adding
Image   system knowledge enrichment



Architecture   Top



How it Works   Top

Within the Knowledge Information Mining (KIM) project and the researches carried out at ETH Zurich and DLR, solutions were prototyped for accessing image datasets through information mining and content-based image retrieval.

A complementary approach is the recognition of objects in images (classical GIS or computer vision problem).

These two approaches are closely related because images have both semantic and visual content.

In KIM, statistical models and machine learning methods were built to "explore and explain" the archived images. They permit the user to associate semantic meanings to (part of) images in free text, as a kind of image language processing. It would be difficult to derive with computer-vision-like methods the semantic content obtained through this process.

In KES, the KIM capabilities of describing images in semantic terms, will be more directly made available to other EO interpretation environments.

A possible scenario sees geographically dispersed users with different skills (or experts in different domains), connected via a simple PC to a KES system, and accessing the pre-loaded images covering a specific region of interest. They can analyse the images and assign meaning to the features they detect. They can also exploit the content of GIS layers or of (even newspaper) articles, and write annotations. They can share among themselves all this information and the growing knowledge, and make them available to other users.
Each user / expert could contribute in terms of:

Image   Auxiliary Data (GIS layers, maps, articles, socio-economical indicators, ...)
Image   Algorithms
Image   Knowledge (image classification or expertise)
Image   New plug-ins to make the tool evolve over time



Output   Top

The result of this project will be a prototype for a generic user interface capable to:

Image   Learn from interactions with users (with different levels of expertise)
Image   Apply the acquired knowledge in the extraction of information from images
Image   Support multi-domain semantic classification of extracted information
Image   Store and retrieve with single queries multi-type data (e.g.: text, GIS layers, images)
Image   Share multi-domain data, information and knowledge


The Final Presentation was held at ESRIN on April 6, 2005.


Contributors to this page: andreadv .

Page last modified on Wednesday 15 of December 2010 15:49:28 CET by andreadv.