51S 
re 
es 
  
  
de Carvalho, Luis 
  
    
(d) (e) 
Figure 7. Lansat TM from (a) 1998 and (b) 1985 and (c) respective difference image. (d) Details of the difference image 
at the first scale level. (e) Smoothed version of the difference image at the fourth scale level. Note the misregistered 
road depicted in (d) while overall differences like phenological condition of vegetation patches are depicted in (e). 
6 DISCUSSION 
The behaviour of changes at different scale levels enables their discrimination according to size classes. Misregistration 
effects and small area changes are depicted as fine details (Figure 7d). Phenological characteristics, atmospheric effects 
and differences in sensor calibration appear in the smooth representation of the signal (Figure 7e). Hence, using 
information from intermediate scale levels one can minimise the problems mentioned above. We found the method less 
sensitive to spatial and radiometric misregistration, although fine details are lost as well. It can be applied to the outputs 
of any change detection technique such as image rationing, principal components, change vector analysis etc. 
Further statistical analysis could also be applied to the wavelet frames but these procedures would be analogous to 
thresholding operations (Ruttimann 1996). The selection of scales to discard and of significant coefficients to keep 
could be driven by statistical tests if no knowledge exists on the size of features of interest. 
Changes are well discriminated but their quantification is not possible when using information from limited scale levels. 
Further research on the combination with other techniques, like region growing algorithms, could be a solution for area 
determination. Applications of the proposed method include, for instance, the automatic selection of changed sites for 
GIS updating. Finally, with respect to geo-information for all, the visualisation of changed sites can be done 
straightforwardly with a simple colour composite avoiding any threshold definition and easily implemented by non- 
experts in image processing. 
ACKNOWLEDGEMENTS 
Thanks to the Brazilian agency CAPES for the Ph.D. research grant of the first author and WWF-Brazil for supporting 
field trips. We are grateful to Thelma Krug at INPE (National Institute for Space Research, Brazil) for providing the 
satellite images and to Newton José Schmidt Prado at CEMIG (Energy Company of Minas Gerais, Brazil) for providing 
the aerial photographs. We also acknowledge with gratitude the valuable help given by José Verdi, Bodinho e Luciana 
Botezelli during fieldwork. 
REFERENCES 
Blanc P., Blus T., Ranchin T., Wald L., Aloisi R., 1998. Using Iterated Rational Filter Banks within the ARSIS Concept 
for Producing 10 m Landsat Multispectral Images. International Journal of Remote Sensing, 19(12), pp. 2331-2343. 
Buiten H., J., Clevers J., G., P., W., 1996. Land Observation by Remote Sensing: Theory and Applications. Gordon and 
Breach Science Publishers, Amterdam, p. 642. 
Burt P. J., Adelson A. E., 1983. The Laplacian Pyramid as a Compact Image Code. IEEE Transactions on 
Comunications 31, 532-540. 
Dai X., Khorram S., 1998. The Effects of Image Misregistration on the Accuracy of Remotely Sensed Change 
Detection. IEEE Transactions on Geosciences and Remote Sensing 36(5), 1566-1577. 
Djamdji J. P., Bijaoui A., Maniére R., 1993. Geometrical Registration of Images: The Multiresolution approach. 
Photogrammetric Engineering and Remote Sensing 59, 645-653. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 345