How
to Introduce Expert Knowledge
in a Classification Process?
The aim of our approach may be understood as correcting wrong classifications
of usual classifiers. So we can consider a two stage process:
- first, an usual classification is performed,
- then a second classification, integrating expert knowledge and out-image
information, is performed in order to improve the previous one.
This approach allows expert knowledge and know-how in satellite image
interpretation to be integrated in a flexible way.
Two approaches may be considered:
- in the first one, the integration of expert knowledge is considered
for updating a previous classification.
It is an updating approach.
- in the second one, the usual classification process is considered as
one source of information and expert knowledge and out-image data are considered
as a second source of information. Then the problem is a fusion
approach of various information sources:
a numerical source and a symbolic source for instance. All fusion processes
may be considered , and we will present a fusion process based on Belief
theory (Dempster-Shafer [Sha81]).
IRIT-UPS
