Peut-on cartographier des taches urbaines à partir d’images Google Earth ?

Abstract

The presented study shows the results of the processing of Google Earth images leading to the delineation of West Africa urban areas with more than 500 000 inhabitants. Since Google Earth images are RGB images with no spectral information, the developed methodology is based on the processing of grey level images in order to retrieve urban areas according to their texture with the help of morphological filters. Images covering some of the studied agglomerations are mosaic images resulting of the composition of satellite images acquired in different conditions. Thus a pre-processing step of image equalization is added in order to reduce the luminance differences and facilitate the extraction. We present here an equalization method based on the « Midway » algorithm, originally developed to standardize the luminance of pairs of stereo images. The challenge here is to adapt the algorithm to be able to treat images with partially different contents. Once the mosaics are equalized, it is possible to use sequences of morphological filters in order to delineate urban areas. The results are compared and validated with the Africapolis vectorial database of urban areas identified by mean of photo interpretation, as well on Google Earth images.

Keywords

• remote sensing
• built environment
• mathematical morphology
• Google Earth
• urban delineation