Advanced Image Processing Tools
for Future Satellite Image
Exploitation Systems
Multi-band /multipolarised SAR data
L and C-band SIR-C data have been used to evaluate relevance for agriculture applications. Data available on a test site, and completed with optical both visible and infrared images have been processed using advanced low-level image processing tools (classification and segmentatin0. Information theory measures allowed to quantify the mutual relevance of these data for classifying fields with different cultures [6-7].
Hyperspectral image data
Future observation satellites may load hyperspectral sensors, ie. Imaging instruments which provide an actual chemical spectrum measurement on each pixel (up to 200 channels with down to 10 nm width). We have simulated such images using AVIRIS data. Sophisticated image processing tools allow to classify the pixels using the spectral information (classification according to the materials observed) [8].
Towards Real-Time Exploitation
The huge image data flow available in multi-satellite receiving stations requires automatic tools for processing, exploiting and disseminating the information towards the end-user. We present here results obtained on three relevant tool families:
-
automatic change detection,
- automatic extraction of high resolution elevation models,
- selective image compression.
Automatic change detection
In many monitoring applications (agriculture, urban areas…), a given site is observed many times. The relevant information is the change between two successive images. Simple methods like image difference are here completely uneffective due eg. To illumination changes or to fine (subpixel) misregistration. We have developed a complete scheme based on fine registration and on structured detection. Moreover, a specific Man Machine Interface has been realized to present first the most relevant changes defected.
Automatic extraction of high resolution Digital Elevation Models
From 98 on, high-resolution optical images should be avaible from commercial programs like
Earthwatch or Space Imaging. Detailed mapping of urban areas is a key application of these data.
Up to now, this application required tedious manual operations due to inage complexity. Fig. 1
presents results of fully atuomatic extraction of high resolution Digital Elevation Models using
two or more inages. This novel methods is based on sophisticated dence correlation tools
developped in our laboratory [9].
Fig. 1 :High Resolution Digital Elevation Model (DEM) obtained automatically from a stereo pari of aerial images (Marseille-France). This zone exhibits roofs and walls which cannot be handled by standard correlation methods used to produce medium resolution DEMs (eg. from 10m SPOT stereo pairs). Here, a highly sophisticated correlator developed by MATRA Systemes et Information has been used (bottom). Comparison with manually derived DEM on the same (Image and reference DEM data : courtesy ISTAR, France)
