ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision", Graz, 2002
parts of the scanner. In this section different mathematical
models are used to handle the geometric errors introduced by
the scanners, the performance of each model is evaluated and
the best model is remarked.
2.1 Two-Dimensions Coordinate Transformations
In order to study and compensate the introduced geometrical
distortions in the scanning process we employed four 2D
coordinate transformation models.
In the first model, Equation (1), three parameters are used to
represent two shifts and a rotation angle between the
hardcopy coordinate system and the scanner coordinate
system. In the second model one more parameter is added to
consider the uniform scale between the two systems,
Equation (2). The third model represents the 4paramter
transformation model. Six parameters are used to represent
two translations, the rotation angle between the two systems,
two different scaling factors, and a skew factor, Equation (3).
The fourth model is the 8paramter transformation model,
Equation (4). A complete study on the 2D transformation
models is discussed in (Mikhail ez. al., 2001).
[X | [cos@ -sin0[.X; X,
e * (D
[Y | |sinO cos0 | Y,
x; cos0 —sin6 || X, Xo
fm SH + (2)
LY» J sind cos0 ||Y Y,
[X,| [ess6 -sime]t o]s. o[x, NES
| Y, le sing cos0 |«& 1110 5, LY Y
(3)
a,X, T Y c
X; aX) T5 Y t1
lr (4)
? a Xi +h Yi +c)
a, X, tb,Y, +1
Where X, Y, — hardcopy coordinates,
XY | = scanner coordinates,
X , Y = shift parameters,
o o
0 = rotation angle,
S — uniform scale factor,
S ,S = non uniform scale factors
Ny
a = skew factor, and
a, a, a, b, b, b, Cp C, = the ¶mter
transformation model parameters.
2.2 Experiment Setup
Some experiments are performed to evaluate the scanning
process as well as to choose the most appropriate model that
minimizes the produced geometric distortions. The process is
based on scanning a precise hardcopy image of known and
accurate point coordinates, Figure 1. The measured
coordinates are used as the reference coordinates for the
hardcopy points. The experiment is performed with a
4400CSE Howled Packard scanner. Two
resolutions were used (2400dpi, and 600dpi).
scanning
qot
++
++
++
++
++
++
++
++
++
++++++++++
++++++++++
++++++++++
++++++++++
++++++++++
+++++++ +++
++++++++++
++++++++ ++
Figure 1. Configuration of Reference Image
The calibration process is summarized in three steps. First
the scanned coordinates of all points are measured digitally.
The points are divided to control points and check points.
The control point coordinates are used to estimate the 2D
transformation parameters for each 2D transformation model.
Then the estimated parameters are used to convert he
scanned coordinates of the check points to the hardcopy
coordinate system. The RMS errors are calculated for each
model and are shown in Figures 2. The Y direction
represents the scanning direction. The results show that the
6-parameter transformation model is the best model to
recompense the geometric errors introduced by the scanners.
Model(1) Model(2) Model(3)
—8— RMS (X), 2400dpi
—$— RMS (X), 600dpi
Figure 2. RMS Errors for the Scanning Process (Pixels)
—&— RMS (Y), 2400dpi
—#— RMS (YY), 600dpi
3. MATHEMATICAL MODELS FOR CLOSE
RANGE PHOTOGRAMMETRY
Due to the wider use of mn-metric cameras for different
photogrammetric purposes such as in registering
archeological buildings, documenting historical writings, and
recording historical buildings, there has been an increasing
demand to calibrate non-metric cameras. Using non-metric
cameras incorporate many problems, including:
1-Non-metric cameras lack a stable inner orientation
technique and irregular methods of calibration.
2-Most non-metric cameras have neither a reseau plate nor
fiducial coordinate system to act as a reference for
photographic measurements.
Model(4)