Page 217 - Digital Analysis of Remotely Sensed Imagery

Image Geometric Rectification 181

GCPs have to be excluded from rectifying an image of a fine spatial
resolution (Table 5.4). The SPOT PAN image that has the finest spatial
resolution among the three is the most difficult to rectify because this
fine spatial resolution is unable to accommodate the inaccuracy of the
GCP coordinates. However, these same coordinates achieved more
accurate rectification for coarser resolution imagery. This is more so if
these coordinates have an uncertainty level comparable to the pixel
size of the imagery being rectified. The accuracy expressed in pixel
size decreases from those shown in Fig. 5.17, the pace of decrease
being larger for images of a finer resolution. If translated into meters,
the relationship identified above is mostly reversed. The highest
accuracy is associated with SPOT PAN and the lowest with Landsat
TM (Gao, 2001). Both SPOT PAN and XL have a very similar accuracy.
The accuracy is noticeably lower for The 30-m Landsat TM because
most GCPs at road intersections are difficult to locate precisely on the
image. Their ease of identification on the 30-m resolution image dis-
appears. The uncertainty in locating the GCPs on the imagery is out-
stripped by the inaccuracy in identifying them. If the resolution con-
tinues to decrease, most of the GCPs will no longer be identifiable on
the image. Its lowest accuracy is attributed to the difficulty in identi-
fying the same set of GCPs on the imagery caused by its low spatial
resolution.

5.6.3 Impact of GCP Quality
Ease of GCP identification varies with the nature of the scene. In the
urban environment, intersections of narrow roads and streets can be
precisely located even on coarse resolution satellite images. The rural
scene dominated by sand dunes lacks distinct landmarks visible on
satellite imagery. There is a limited choice in selecting quality GCPs,
which are located less accurately on the topographic map and the
satellite imagery than their urban counterparts. Unreliable GCPs had
to be selected to make up a number comparable to that of the urban
scene for the purpose of comparison.
The impact of the nature of the scene on image rectification accu-
racy is illustrated in Fig. 5.18 in which the numbers of GCPs were
standardized to percentages of all GCPs to compensate for the effect
of the varying number of GCPs used for the urban and rural scenes.
The ease of identifying GCPs on the ground and on the satellite
image exerts a direct impact on rectification accuracy for SPOT PAN
(Fig. 5.18a). At any given level of GCPs the rural scene is rectified less
accurately than its urban counterpart if the image has a fine spatial
resolution. When all GCPs are used, the urban image is rectified much
more accurately than its rural counterpart. However, the gap in recti-
fication accuracy is gradually bridged as more and more poor GCPs
are excluded from rectification. Eventually, the difference between
the two scenes in accuracy nearly disappears when half of the GCPs