Image Processing Reference
In-Depth Information
However the change of resolution is well shown and would clearly impact photo-
interpretation. Figure
11.4
shows the result of our smoothed encrustation procedure
in which the visual differences are small. However they still exist. Close examina-
tion shows that the outer periphery exhibits a larger variability in Fig.
11.4
than in
Fig.
11.2
. This higher variability makes a smoother transition from the high resolu-
tion part to the lower resolution portion of the scene. The shapes in the periphery
are also more visible. For example, note the white shape at the upper left corner of
the imagette: its contours are less blurred in Fig.
11.4
than in Fig.
11.2
. The same
is true for several features along the upper periphery of the imagette. Though the
differences are faint, it visually confirms the better results achieved by our proposed
encrustation method. The border has been smoothed outside the imagette area with-
out the alteration of the features larger than 5.6 m.
The disturbing effect of the discontinuity along the periphery of the imagette can
be quantified by smoothing the two encrusted images in order to have a uniform
effective resolution of 5.6 m across the image scene including the higher resolution
imagette. These degraded images are compared to the 5.6 m smoothed image. In
the inner and outer periphery, the difference in variance between the encrusted
images and the 5.6 smoothed image account for about 22% of the variance of the
latter image. On a pixel-by-pixel basis, 50% of the pixels have a similar value in the
smooth encrusted and the 5.6 m smoothed images, plus or minus 5 digital numbers
(images are coded in 16 bits). The better results demonstrate quantitatively the
attenuation of the edge effects by the proposed method.
A second quantified test deals with the accuracy of the information of small sizes
injected in the outer periphery in order to decrease discontinuity. The results indicate
that our proposed encrusted image better represents the 0.7 m reference image than
the raw encrustation method does. In order to assess the quality of the simulation
of the actual information in the reference image, structures of any size down to 0.7 m
were examined. The differences between the reference and the encrusted images
were computed for the outer periphery. As expected, the new approach provides
better results than the raw encrustation, i.e. closer to the ideal values. For example, the
difference of variances decreases from 63% in the raw encrustation approach down
to 59% for our smooth encrustation method (ideally, this difference should be null),
with a correlation coefficient equals 0.87. This demonstrates that the new approach
provides a better simulation of the actual information than the other method does.
11.4
The ARSIS Concept for Fusion of Images in Urban Areas
11.4.1
Introduction
Several studies have documented the utility of merging broadband higher spatial
resolution images with imagery that are low in spatial resolution imagery but high
in spectral resolution. Many methods have been developed for that purpose, which
