534 
DISCUSSION 
From the statistical aspects, the resampling via CC seems 
to be more accurate compared with other schemes, whereas 
other approximate interpolation merits condideration to 
produce the radiometric values in the geometrically cor 
rected images. Three approaches as NN,BI, and CC are used 
to study the radiometric fidelity of images. It is seen that 
the overall statistics of classification is only slightly 
affected by geometric correction, whereas the pixel-by 
pixel differences between the classification of uncorrected 
and resampled data are noticeable for a few land use classes 
from the aspects of radiometric fidelity. Furthermore, it 
seems that such tendency does not depend on the atmospheric 
scattering effects on images. 
In conclusion, it seems that from the aspect of radiome 
tric fidelity the classification map via NN resembles well 
to that of the uncorredted data compared with the classifi 
cation maps via BI and CC, which show some peculiarity for 
a few training sites. Furthermore, such a situation does 
not change even after removal of the atmospheric effect on 
classification maps. On the other hand, the overall statis 
tic of class occupancy is almost negligibly affected by 
the geometric correction. 
Finally, we thank Fr. K.Takemura for his numerical computa 
tion. 
Ahern,F.F., and Murphy,J. 1978, Radiometric calibration and correction 
of Landsat 1,2, and 3 MSS data.R.R.78-4, C.C.R.S. 
Bellman,R., and Kalaba,R. 1965, Quasi 1inearization and Nonlinear Boundary- 
Value Problems. American Elsevier Publ.Co.Ltd., New York. 
Bernstein,R. 1976, Digital image processing of Earth observation sensor 
data. IBM. J.Res.&.Develop.,Vol.20, pp.40-57. 
Busbridae, Ida.W. 1960, The Mathematics of Radiative Transfer.Cambridge