iY 
st digital 
the most 
t recover 
Jests that 
those pixels 
lane of the 
ammetry to 
! placed on 
otter. Why 
librations if 
ner should 
ical plotters 
scted under 
s. If the 
e analytical 
apts can be 
radiometric 
in an image 
sume that 
el, we get a 
us that our 
cantly less 
Its and not 
1 pixel scan 
ources 
arily results 
are several 
) impact on 
are derived 
e tension to 
nbined with 
n focus and 
id. Flat bed 
er plates to 
s an easier 
y. Flat bed 
ceptible to 
rs designed 
signed with 
the graphic 
m scanners 
cient for the 
arp JX-600 
a flat cover 
age drive or 
is in the 80 
reasonably 
grammetry, 
Some scanners offer calibration procedures using film 
or glass plate grids that can be measured. Some 
scanners state their positional accuracy, many do not. 
Some users will perform an affine transformation on the 
4 or 8 fiducials of a diapositive to see how large the 
residuals are. This is sometimes misused to judge the 
accuracy of the scanning stage. Due to severe 
limitations on redundancy, and a lack of calibration 
points within the main scanning area of the film, this 
method should not be used to judge scanner accuracy. 
The following is a partial list of error sources influencing 
the geometric accuracy of a scanned digital image: 
X-Y stage or drum positioning: Is it repeatable? 
Can it be calibrated?, Does it change with 
temperature? 
Film flatness: Does it stay flat? Is it optically 
flat? Can it bubble? 
Radiometry: Are the pixels highly correlated? 
Is there a lot of noise? Is it out of focus? Is 
there enough dynamic range? 
Lens distortion: Is it significant? Can it be 
calibrated? 
Interior Orientation: Can it be performed? What 
is the measuring method? 
Sensor Errors: Is the camera digital output 
linear? Can it be calibrated? Are the pixels 
square? 
A manufacturer is not in a position to guarantee the 
accuracy of a scanned image. Too many operational 
factors come into play. The supplier of photogrammetric 
products is generally responsible for accuracy 
guarantees. The user of the scanner must be able to 
produce the accuracy statement. The manufacturer 
can supply the means by which the user can test the 
image accuracy. 
A standard for stating the geometric accuracy for a 
scanned image, and a standard set of methods for 
achieving the statement of accuracy would be a big 
help. 
Calibration Methods 
Some scanner stages make routine use of calibration 
procedures to indicate and prove their geometric 
fidelity. These calibrations can be used at any time to 
analyze or set the geometric accuracy. Scanners that 
provide this knowledge, allow the user to calculate the 
scanners effect on C-factor or accuracy. The Helava 
DSW 100 and 200, The Zeiss-Intergraph PS-1, and 
others use this type of calibration to verify geometric 
quality. The Vexcell VX 3000 scanner accomplishes the 
same thing by providing grid calibration with each 
scanned image (Leberl 1992). Stable base films with 
calibrated grid lines or a glass grid plate are the most 
common methods of checking and calibrating X-Y 
stages. This is a fairly simple procedure, it gives 
confidence to the user and quality statements for the 
product. 
Besides the stage, the resulting pixel data needs to be 
checked for geometry. An accurate stage will not 
assure an accurate motion of the stage with-respect-to 
the sensor or vice-versa. The relationship of the sensor 
to the camera, and the camera to the stage needs 
checking. This can be short circuited by testing the 
resulting image. One method of doing this is by 
135 
scanning a calibrated grid and then measuring the 
resulting image. 
EXAMPLE CALIBRATIONS OF THE DSW 200 
The DSW 200 is a photogrammetric film image scanner 
produced by Helava Associates Inc. Like the DSW 100, 
this scanner uses a 10 X 10 inch flat bed stage with 1 
micron encoders. It typically calibrates about 2 microns 
Root Mean Square (RMS) error in X,Y position. It Scans 
an image using a 2k X 2k area array sensor at physical 
pixel sizes between 7 and 15 microns. It outputs pixels 
to the host computer at approximately 1 million pixels 
per second. The pixels for each area (about 2kX2k) are 
tiled together by the host computer to generate a 
seamless image of up to 10 by 10 inches. 
Scanning Mechanics 
The X-Y stage is commanded to move to a given 
position above the sensor and comes to rest before an 
image is acquired. A highly uniform light source is 
transmitted through the cover plate, through the film, 
through the base plate, through an imaging lens, and to 
the sensor elements in the camera as shown in the 
figure below. 
Light Source-------------------------- > NG 
20 \D 
Cover plate (glass)----> 
Film to be scanned------ > 
Base plate (glass)-----> 
  
EE EE OSPR. 
  
I 
i 
Lens--0. ALZA. EMI > ©O 
Sensor ----------------------------- > 
  
  
By calibrating the relationship of the sensor array to the 
stage coordinate system, we can compute each 
position of the stage for collecting an entire array of 2k 
X 2k images to cover the entire stage area. Each stage 
position is accurate to less than 3 microns RMS, and 
each pixel position within the sensor is accurate to less 
than 3 microns RMS. How do we know this? 
Stage Calibration 
The X-Y stage positioning of the scanner is calibrated 
using an etched glass grid plate. It is on this grid as well 
as the precision and high stability of the opto- 
mechanics that stage accuracy is based. The grid lines 
etched on the grid plate are accurate to less than 1 
micron as guaranteed by the manufacturer. This grid 
plate has a grid spacing of 20 mm in X and Y. Based on 
this calibrated data, DSW 200 will automatically find 
every grid intersection, measure it, and record the 
physical stage coordinates to the nearest encoder 
micron. Image processing algorithms are used to find 
and measure the intersection to stage precision. 
A table lookup of stage corrections can now be built 
between the physical (encoder) stage coordinates and