One can analyse urban areas through remote sensing. This analysis is very useful in many fields:

- Update maps
- Observe urban populations
- Set up radiomobile networks
- Analyze urban areas evolution

One often study urban areas by using the following data:

- Optic satellite images (SPOT, LANDSAT)
- Radar satellite images (ERS, JERS, Radarsat)
- Aerial images

One study urban areas from two point of view in order to:

- Extract a urban mask.
- Distinguish different regions inside the city itself.

Firstly we aim at precisely __defining a urban mask from high
resolution satellite images__ (SPOT5). You cannot extract
urban areas from satellite images through only grey level
information. Hence some approaches use mathematical
morphology. Others approaches use classical texture parameters
(coocurrence matrix, vegetation index...). In this work we model
the image by a Markov randon field.

The steps are the following:

- Analyze the texture of urban areas through a Markovian modelling. We defined a new textural parameter derived from a Markovian Gaussian model. This new parameter takes into account the variance of the image in eight directions.
- Classify the image of parameter by a modified fuzzy Cmeans algorithm. This algorithm include an entropy term. The number of classes does not need to be known a priori.
- Segment and regularize through Markovian modelling
- Eliminate false alarms by using an image of markers.
- Some results
Last modified: Wed Feb 18 16:59:44 MET 2004