ON THE AUTOMATED ASSESSMENT OF GEOMETRIC SCANNER ACCURACY 
R. Seywald 
Institut für Computerunterstützte Geometrie und Graphik 
Technical University Graz 
Münzgrabenstr. 11, A-8010 Graz, Austria 
seywald@icg.tu-graz.ac.at 
Commision |, Working Group 5 
KEY WORDS: Scanner, Accuracy, Automation, Analysis, Geometry 
ABSTRACT 
A system is presented to investigate the geometric accuracy of scanners. Evaluation procedures and algorithms are described 
to analyze various types of scanners, from the low cost DTP scanner to the high performance film scanner. Evaluation can be 
based on a test target which allows both, manual visual analysis as well as automated algorithm evaluation. An assessment of 
quality has to differentiate between local and global geometric accuracy. Experimental evaluation results were obtained with 
scanners using different scanning principles. Further the requirements of a reliable geometric accuracy test target are discussed. 
KURZFASSUNG 
Ein System zur automatischen Evaluierung der geometrischen Auflösung von Scannern wird präsentiert. Die vorgeschlagenen 
Evaluierungsprozeduren und Algorithmen ermöglichen die Untersuchung verschiedener Scannertypen, vom DTP Scanner bis 
zum hochgenauen Filmscanner. Die Evaluierung, welche auf einem Testmuster basiert, kann sowohl manuell visuell als auch 
automatisch erfolgen. Bei jeder Untersuchung muß zwischen globaler und lokaler geometrischer Genauigkeit unterschieden 
werden. Experimentuelle Resultate zeigen die Anwendbarkeit der vorgeschlagenen Methode zur Untersuchung unterschiedlicher 
Scanner. Weiters werden Überlegungen zum Entwurf und zur Anwendung von Geometrietestmustern diskutiert. 
1 INTRODUCTION accuracy evaluation well known measuring mark positional 
values (x(i), y(i)) of the underlying analog target need to 
be compared with the values (x(i)’, y(i)’) obtained from the 
digitized image. A linear conformal or affine transformation 
of points has to be performed to relate the two co-ordinate 
systems to one another. Residual differences in x- and y- 
directions can now be used to calculate the absolute, relative 
global and local values of accuracy. The absolute errror is 
expressed in pixels or mm. Relative accuracy is obtained by 
relating the error of the distance between two marks to the 
distance itself. 
Among other possibilities digitization by scanner is the most 
important way to obtain digital data from an analog image. 
Geometric resolution, geometric accuracy, radiometric reso- 
lution, colour reproduction and speed are topics which have 
to be considered in an assessment of quality. The geomet- 
ric accuracy of a scanner is defined by its ability to create a 
geometrically exact reproduction of an analog image. Con- 
sequently the geometric accuracy is a measure which shows 
how accurate the geometric proportions of the original analog 
Image are maintained in its digitized copy. 
This work describes procedures to analyze the geometric ac- 
curacy for variuos types of scanners, but in particular for high 3 SCANNING PRINCIPLES 
performance film scanners. The evaluation can be based on 
well defined high accuracy test targets and automated algo- Image scanners can be classified according to their main tech- 
rithms using image processing. Specific algorithms have been nologies. With regard to geometric accuracy basically one 
implemented in a software system called SCANEVAL in order has to distinguish between flatbed and drum scanners. The 
to achieve an objective, operator independent, easy to use, configuration of the built-in photo-detector is an additional 
time saving and comparable process of geometric accuracy criterion to characterize scanners. Image sensor technologies 
evaluation. Results may be used for a thorough comparison [Baltsavias, 1994c] are : 
between different scanners as well as for detection of long - point sensors, 
term stability; they may also serve as a basis for geometric - linear array sensors, 
scanner calibration. - Square array sensors. 
For geometrically high accurate scanning the fled-bed princi- 
2 BACKGROUND ple currently is the the preferred technology. 
The number of tiles or swaths to be sewed together to one 
The geometric accuracy can be assesed by use of geomet- scan depends highly on the dimension of scanners the photo- 
ric accuracy test targets [Baltsavias, 1994b] [Seywald, 1994] detector. Inside an individual scanning swath sources of error 
[Simonis,1991]. The targets contain highly symmetric mea- are lens distortion, lack of focus, CCD noise, rotation and 
suring marks distributed over the whole measuring area, translation of the sensor and lighting errors. Using linear ar- 
where the mark's positions and distances between them are ray technology large images at contemporary high resolutions 
well defined. Local geometric accuracy is which is measured can be achieved by scanning the image with multiple swaths 
by means of marks in a relatively small area (e.g. 2cm x and stitching adjacent swaths together. Scanning with square 
2cm), whereas global geometric accuracy is measured across array sensors [Leberl, 1992] may require the assemblage of 
the entire scanning area (e.g. 20cm x 27cm). For geometric neighboring tiles geometrically and radiometrically exact. In 
182 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996 
the: 
due 
Sin: 
ing 
for 
deli 
errc 
Gec 
peri 
dpi 
film 
abl 
is p 
met 
as \ 
get 
me: 
int 
can 
buil 
of « 
fror 
to 
Der 
on. 
Wh 
sibl 
the 
test 
me: 
as 
typ 
Ho 
The 
film 
atu 
Nor 
tem 
Ing 
tior 
equ 
sigr 
Fig 
tize 
qua 
arr( 
hai 
on 
ric 
per 
squ