Spatial enhancement of multispectral images on urban areas - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 143

one third (88) of the blocks is flagged. It is interesting to link the
results of Fig. 10.5 with Table 10.1. Since in Table 10.1 GLP-CA
is the best algorithm in terms of average D s errors, followed
by GSA-CA and GMMSE, it is evident that the distribution of
D s errors is different for the three algorithms. In particular, the
distribution of errors for GMMSE is flatter than that of the other
methods. This fact may indicate that the algorithm exhibits a
behavior that is more stable with respect to landscape variations,
which may occur frequently in an urban scenario.
Results relative to IKONOS images are reported in Fig. 10.6.
The order by which the images are presented is the same as
in Fig. 10.4. GS method appears more sharpen than the others
on vegetated areas because of a heavy and excessive injection of
Pan details in the blue and green bands. GLP-CA and GSA-CA
colors are quite similar to the expanded MS on vegetated areas

due to the local injection model. GMMSE colors are very similar
to the EXP image. On urban areas all the algorithms properly
inject spatial details. Only a careful analysis reveals a little blur on
GLP-CA and GMMSE contours because of the misregistration
problem outlined above.
Figure 10.7 reports the maps of the details that each method
injects into the expanded multispectral images with the same
order of Fig. 10.5. Themaps are obtainedwith the same procedure
described for Fig. 10.5. GMMSE and GSA-CA obtain the best
results. Only one block is flagged. Also GLP-CA is efficient and
only two blocks are flagged. Conversely a higher number of
blocks is flagged by the GS algorithm (15).
By analyzing these results with the scores reported in
Table 10.2 and in Table 10.1, it appears that for the IKONOS
image the distribution of D s errors is extremely flat and stable.

(a) Gram-Schmidt (b) GLP-CA

(c) GSA-CA (d) GMMSE

FIGURE 10.7 Maps of the details injected by each method into the expanded MS bands of the IKONOS scene. Details are
represented in full color for the red, green and blue bands. The blocks of size 32 × 32, on which the spatial distortion D s
exceeds a threshold value of 0.3, have been superimposed.