John Bosco Kyalo Kiema 
  
  
  
  
  
  
20 —- n : 
3 x 
- 18 » > UP cs 
x N Aa, 
= N ro 
2 18 d 
14 
12 
10 
5 18 15 20 3G 50 70 100 
Compression Rate 
Figure 6: Relationship between PSNR and compression rate 
5 SUMMARY AND CONCLUSIONS 
This study underlines the importance of multi-sensor data fusion in the classification of urban environments. In 
particular, the need to integrate multispectral and geometric datasets is underscored. The need to incorporate context 
information in image segmentation is also highlighted. This is achieved by expanding the object feature base in order to 
exploit both spectral and spatial feature characteristics. Through this, higher classificatiori accuracy and better semantic 
differentiation between the various object features can be achieved. 
Data compression is conventionally employed in remote sensing within the context of data transmission and storage. 
However, the examination of the influence of this on the quality and further processing of satellite sensor imagery 
represents an important research and development topic. This is basically because of the existing dilemma between the 
huge amount of data often captured by remote sensing sensors and the technical restrictions in using this. Compression 
rates greater than 10 have been proposed for the next generation of commercial sensors in view of their higher spatial 
resolution and larger swath widths (Fritz, 1997). In this regard, lossy compression schemes provide the only viable 
solution. The superiority of wavelet-based methods over the standard JPEG technique in the compression of remotely 
sensed data has been demonstrated (Shiewe, 1998). Nonetheless, the smoothing effect of wavelet compression, 
especially at higher compression rates (K 2 50) is undesirable. 
The compression rate beyond which the smoothing effect becomes evident represents the critical compression rate that 
defines the range within which the compressed imagery can be used in further processing e.g., classification. It is clear 
from the results obtained in this study that a compression rate of up to 20 can be adopted for the classification of urban 
environments using Daedalus ATM imagery fused with airborne laser scanning data without adversely affecting the 
classification results. Further studies on this are still required to determine: (1) whether the cut-off value is influenced 
by the spatial resolution of the imagery, and (2) whether this depends on the scale at which the actual variation exists or 
on that which the user is interested in. 
ACKNOWLEDGMENTS 
The author wishes to acknowledge the support of Prof. Dr.-Ing H.-P. Bihr and the Deutsche Akademischer 
Austauschdienst (DAAD) in the preparation of this paper. 
REFERENCES 
Bähr, H.-P., and Vógtle, T., 1998. Digitale Bildverarbeitung: Anwendung in Photogrammetrie, Kartographie und 
Fernerkundung. Wichmann Verlag. Heidelberg., pp. 220-228. 
Báhr, H.-P., and Vógtle, T., 1999. GIS for Environmental Monitoring. E. Schweizerbart'sche Verlagsbuchhandlung, 
Stuttgart. 
Chui, C. K., 1996. Introduction to Wavelets, Academic Press, San Diego. 
  
494 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.