 |
 |
   |
|
|
|
Landsat Image
These approaches are generally based on a choice of classes and all pixels must be aggregated to one of these classes.
The expert usually gives samples of each searched class. These samples are shown interactively on the image, by means, for example, of connected sets of pixels drawn with a mouse on the screen of a computer. At the end are obtained, for each class, connected sets of pixels on the image which may be considered globally as the "heart" of this class.
All classical methods take into account these samples and for each pixel compute its "distance to each class" or the belonging probability or certainty factor to each class. Finally a decision process (generally based on a maximum criteria) decides for each pixel the best class label.
For example:
A Landsat-MSS image of the Palni Hills in India is given. A French CNRS laboratory in Toulouse ("Carte internationale du tapis végétal") was interested to obtain the vegetation cartography by means of this satellite imagery and an expert was chosen for that particular task. The usual computer based help was of course thought as "data analysis" process: classical classifiers. In the Landsat-MSS image, an area of 400x400 pixels has been extracted, as shown in the following pages.
IRIT-UPS