similarity transformation and the standard deviations are
reported at three different scales in Table 4.
Table 4. Standard Deviation Between Two Operators
Film Scale Tube Scale Ground Scale
Imag
x = 8.4 ym 3.6 ym 14 m (30
y = 9.7 4.2 17
R = 12.8 5.5 22
Imag
It is apparent that image identification and measurement (31
contributed a significantly larger variability (R = 22m)
than map identification (max = 10m).
Imag
Image Adjustment to Ground Control. The coordinates of (15
each of the four images were adjusted to ground control
with each of the four transformations in turn. Table 5
gives the resulting standard deviations. Note in the Imag
table that x = y for the perspective and projective trans- (14
formations. This results from the computer program
(Bender, 1970) pooling the x and y residuals when
calculating the standard deviation.
a ge
The similarity and affine transformations, being two- 0 f t
dimensional, utilize UTM coordinates directly as input
ground coordinates; the perspective and projective trans- Sele
formations utilize space rectangular coordinates. Table 5 rbv
indicates that residuals whose resultant standard deviation usin
is between 90 to 100 meters can be expected from the prec
similarity transformation and between 85 to 90 meters for to t
the affine. By taking into account the three dimension- and
ality of the ground control (perspective and projective seen
transformations), the values drop down to between 30 and seen
45 meters. Also, the two-dimensional transformations give expo
residuals in the y direction which are approximately equal the
to those in the x direction. foun
cat i
row
MOSAICS were
norm
In the experiments with Landsat 3 RBV imagery as reported prod
in the preceding section, horizontal control points were than
identified on 1:24,000-scale paper maps and visually the
transferred to a much smaller 1:500,000-scale RBV image. havi
seal
A possible improvement in geometrically controlling the idea
RBV imagery is to use control which has been field
identified by aerial photographs instead of using map Cont
identified control. It is anticipated that more accurate Aeri
positioning of control points onto the RBV imagery could aero
be accomplished by direct transfer from field controlled Nant
photographs, thus, effectively eliminating the error due cont
to the visual transfer from the paper maps. In order to feet
verify this expectation, a Landsat 3 RBV mosaicked image Cent
base map of Cape Cod, Massachusetts, was controlled through this
the use of aerial photographs. Landsat 3 RBV images were prob
combined into a single mosaic as a 1:100,000-scale base for tria
adju