the calibrated grid plate coordinates. This lookup table 
of corrections is now used to produce calibrated stage 
coordinates for any commanded stage position. This 
calibration can be performed at any time and takes 
about 30 minutes. Additionally, there is a "stage 
verification" mode which measures a subset of the grid 
intersections and tells the user if there is any problem 
with the current calibration. This procedure takes about 
5 minutes. The verification compares the calibrated 
stage coordinates against the actual grid plate 
coordinates and shows the user the RMS error in X and 
Y. If this exceeds their desired threshold, the stage is 
recalibrated. The stage verification procedure is 
recommended about once a month and recalibration is 
rarely necessary. Due to a well proven design, the 
stage holds calibration for very long periods of time. 
Sensor Geometric Calibration 
The sensor image of the DSW 200 is also calibrated for 
geometric accuracy to the same level as the stage. 
This can be done by another glass grid plate that has a 
grid line spacing of 5 mm and whose intersections are 
also accurate to 1 micron. A 6 by 6 grid of intersections 
will then appear on the focal plane of the 2k by 2k area 
array when using an optical pixel size of 12.5 microns. 
This grid data again permits us to automatically find and 
measure each grid intersection to sub-pixel accuracy 
using image processing. A geometric transformation 
between the image pixel coordinate system and the 
calibrated grid coordinate system in computed. This 
transformation only allows uniform scale and rotation 
between the sensor and grid. Deviations from this ideal 
geometry are computed and output for display to the 
user. No more than 3 microns rms is tolerated. This is 
performed as a check on the setup of the camera to lens 
to stage optical path. If there are any alignment errors in 
the stages, lenses, or array, they are discovered during 
this procedure. This verifies the camera and lens setup 
for each new scanner. 
Checking The Final Image 
The scanned image is an array of pixels residing in the 
computer or on the hard disk. The image can be 
displayed and measured. lf we assume a maximum 
scan of interest to cover about 220mm x 220 mm (9 inch 
x 9 inch), then we should have a reference target of that 
size to scan. A calibrated grid plate or film transparency 
seems ideal, with intersections at a spacing sufficient to 
uncover local distortion but not so frequent as to be 
impractical to measure. The technology to produce such 
accurate grids is readily available so no new wheel 
needs inventing. This proposed reference would be 
scanned and an image file produced. Next in this case, 
every grid intersection would be measured by the human 
placing a cursor on each perceived grid intersection or 
by using image processing. When using the human and 
a computer monitor, we would typically zoom the image 
by a factor of 2 or 4 to help place the cursor to sub-pixel 
positions on the grid intersections. However, an 
algorithmic processing method would be preferred due 
to its ease, speed and repeatability if one could be 
agreed upon. If universally agreed upon algorithms were 
employed, then a real scanner standard would begin to 
be possible. 
136 
Let us assume the grid has intersections every 20 
millimeters. That would require measuring about 100 grid 
intersections over the entire image. With these 
measurements, a 4 parameter transformation from 
pixels to grid coordinates could be computed. This 
transformation physically models rotation, translation, 
and a scale change between the grid coordinate system 
and the pixel coordinate system. The RMS of the 
residuals from this transformation is a good indicator of 
accuracy of the scanned image because the 
redundancy is so high and testing was done throughout 
the scanned image. This procedure can be performed 
by us the manufacturer, or by the user. This permits a 
statistically rigorous check on the resulting image 
geometry directly. 
RECOMMENDATIONS 
Since the user of the image scanner or scanned image 
data is really in control of the geometry of the image, 
and since it is a relatively easy procedure to verify the 
quality of a scanned image, we would recommend that 
the user of the scanner test and produce a standard 
accuracy statement. We would also recommend that 
the manufacturer of the scanner provide the software 
tools necessary to measure a calibration plate or film so 
the user can produce the accuracy statement. We 
would recommend that the professional associations 
establish the specific methods by which this should be 
done and the minimum content of the statement. The 
manufacture or the user could supply the calibration film 
or plate. A large 9 by 9 inch scan should be checked 
with a minimum of 20 points spread throughout the scan 
area. This would permit sufficient redundancy when 
checking the quality using a 4 parameter 
transformation. The resulting accuracy statement 
should minimally state the RMS in X and Y directions 
along with the maximum fit residual. 
References 
Leberl, Franz W., 1992, Precision Scanning Of Aerial 
Photogaphy: ASPRS Technical Papers, 1992 ASPRS- 
ACSM Annual Convention, Vol.1 pp. 247-252. 
Mikhail, Dr. Edward M., 1992, Quality Of 
Photogrammetric Products From Digitized Frame 
Photography: 1992 ISPRS, International Archives Of 
Photogrammetry And Remote Sensing, Vol. XXIX, Part 
B2, Commission ll, pp. 390-396. 
Sarjakoski, Tapani, 1992, Suitability Of The Sharp JX- 
600 Desktop Scanner for The Digitization of Aerial Color 
Photographs: 1992 ISPRS, International Archives Of 
Photogrammetry And Remote Sensing, Vol. XXIX, Part 
B2, Commission ll, pp. 79-86. 
  
« 
Presque 
produit 
plupart 
Le bala 
détermi 
de tra 
balayeu 
l'étalc 
soit ét 
Mots-cl