Barsi, Arpad 
  
The first experiment was the usual classification with the 
original image bands. Then PCA-analyzed and transformed 
bands were processed. The results of these classifications are 
very similar as Figure 15 demonstrates. The map is very 
smooth; it contains too much urban pixels. The redundancy 
could be reduced in this way, which accelerated the design and 
training. 
The principal component analysis and Karhunen-Loeve 
transformation are nice tools for data reduction. 
Considering pixels’ neighborhood information just the opposite 
results have arisen. The thematic map is “noisy”, disturbing 
pixel differences could be noticed. The design is much slower 
and harder, than the original version. The application of the 
“raw” direct and indirect neighborhood has therefore not too 
much sense. 
The main result of current project is the combination of the 
neighborhood information and the PCA data compression 
technique. The resulting thematic map has enough details, but is 
generalized. The output of this last method is high quality, 
esthetic thematic map. The 4- or 8-neighborhood are differing in 
the designing, training and processing time; the map quality is 
very similar. 8-neighborhood map has more contrast (“edge 
detected”), but the cost-benefit analysis would prefer the 
simpler and faster 4-neighbor solution. 
Current research work is just documented, isn’t at the final 
stage. Future works are planned in the application of PCA after 
the training set preparation, in the extension of neighborhood 
and evaluating information content series. 
  
REFERENCES 
Brause, R., 1995. Neuronale Netze, B.G. Teubner, Stuttgart 
  
Erdas1994, Erdas Field Guide, Erdas Inc., Atlanta 
Image1998, Image Processing Toolbox Version 2.1 User’s 
Guide, MathWorks Inc., Natick 
Neural1997, Neural Network Toolbox Version 3.0 User's 
Guide, MathWorks Inc., Natick 
Matlab1998, Using Matlab Version 5.2, MathWorks Inc., 
Natick 
Paola, J.D., Schowengerdt, R.A., 1995. A detailed comparison 
of backpropagation neural network and maximum likelihood 
classifiers for urban land use classification, TGARS, Vol. 33, 
No.4, pp. 981-996 
Rojas R., 1993. Theorie der neuronalen Netze — Eine 
systematische Einführung, Springer Verlag, Berlin 
fa SARL, Bal A SNE 22d 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 147