Page 219 - Digital Analysis of Remotely Sensed Imagery

Image Geometric Rectification 183

are excluded. On average, the difficulty of identifying GCPs on the
ground degrades rectification accuracy by up to one pixel (0.85 pix-
els larger for SPOT PAN).
The above finding still holds true for the SPOT XL image
(Fig. 5.18b). For instance, the urban scene is more accurately recti-
fied at all levels of GCPs with the average residual being 1.09
lower. Besides, a much higher proportion of the selected GCPs are
useable in achieving an acceptable level of rectification accuracy
for the urban scene. This relationship is reversed once the percent-
age of GCPs used drops to 90, this percentage drops to 50 for the
rural scene due to the difficulty in locating these points accurately.
The larger uncertainty surrounding the locating of these points
makes an accurate rectification harder to achieve.
As with the previous two types of SPOT imagery, the accu-
racy is also higher for Landsat TM imagery when all GCPs are
retained, even though neither accuracy is acceptable (Fig. 5.18c).
However, the difference is almost nonexistent after a few GCPs
are removed. However, this relationship is reversed once the per-
centage of GCPs used drops to 90, in sharp contrast with the pre-
vious two images. Consequently, rectification accuracy is actually
higher for the rural scene. The difficulty of identifying GCPs
becomes less significant if the image being rectified has a coarse
resolution. In this case the nature of the scene exerts little influ-
ence on rectification accuracy. Owing to the image’s coarser reso-
lution a larger portion of the selected GCPs are useable to achieve
the required accuracy.

5.7 Image Orthorectification
The image georeferencing covered so far has concentrated on the
horizontal position (E, N) of pixels in the output image while their
elevation (H) on the ground has not been given any consideration.
This practice is acceptable with spaceborne satellite imagery of a
coarse to medium spatial resolution in which topographic relief-
induced shift in pixel position is negligible, or in applications in
which precise geographic location is not a primary concern. In local
applications (e.g., urban planning) involving very high spatial reso-
lution satellite imagery, the geometric position of pixels needs to be
determined accurately as well. Such image georeferencing must take
into consideration the minor shift in pixel position caused by topo-
graphic relief. This brings out the issue of image orthorectification.
Topics covered in this section include differences between ortho-
graphic and perspective projections, and the methods and procedures
of image orthorectification.