The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
To be valuable, this representation has to report accurately the
quality of the image and the performance that one can expect
from a particular application whatever the image is. However,
most degradation effects are sensitive to the image. It is easy to
understand that a spatial smoothing will have a greater influence
on a city image than on a uniform area (Fig. 7).
MAE
Figure 5: Quality for different values of white noise on three dif
ferent images.
MAE
Figure 6: Quality for different rates for JPEG 2000 on three dif
ferent images.
When considering spectral smoothing, the same conclusion holds.
Moffett3 and harvardl images have more high frequency compo
nents than mojfett4. The spectral content of harvardl is com
pletely different from moffettS and mojfett4. Thus, the quality
figures differ (Fig. 8). However, as we can expect the influence
of the smoothing on applications to be different also, this is not
surprising.
It is important to notice that the representation is quite robust for
the three different images moffett3, mojfett4 et harvardl for the
white noise and JPEG 2000 degradations (Figs. 5 and 6).
3 QUANTITATIVE EVALUATION
3.1 Distance between degradations
The ideal situation to reach, the Grail of the quality criteria would
be to infer precisely the degradation impact on application know-
MAE
Figure 7: Quality for different values of spatial smoothing on
three different images.
MAE
Figure 8: Quality for different values of spectral smoothing on
three different images.