Context and objectives
The subject of this research is the development of image processing software for the analysis of remotely sensed data. In this domain methods and techniques were developed for statistical classification, for texture analysis and for multi spectral edge detection.
Expected scientific results
- statistically significant improvements of up to 4 of the global accuracy were observed, by using robust statistics.
- by using texture features derived from panchromatic SPOT images of agricultural scenes as additional bands in a classification together with spectral features, as derived from an XS-SPOT image, a 20 % increase of the global accuracy was observed, as compared to the purely spectral classification.
- Two methods containing the complementary edge information yield more accurately localised edges and more details. The first method computes a multi spectral gradient with a magnitude equal to the maximum of the separate greytone gradient magnitudes, and with the same direction as the corresponding uni-spectral gradient. The second method computes a theoretically correct multi spectral gradient as a vector in each point of the space of p-vectors (for p-spectral bands), in the direction in which the change of the spectral values is greatest.