COLOR CALIBRATION OF SCANNERS
USING POLYNOMIAL TRANSFORMATION c
V
Ibrahim Yilmaz*, I.Oztug Bildirici ? Murat Yakar " Ferruh Yildiz e
8]
* Afyon Kocatepe University, Faculty of Engineering, 03200 Afyon, Turkey — iyilmaz(@aku.edu.tr e
? Selcuk University, Faculty of Engineering and Architecture, Department of Geodesy & Photogrammetry Engineering, M
42075 Konya, Turkey — (bildirici, yakar, yildiz)@selcuk.edu.tr p
a
Commission V, WG V/1 S
T
fc
KEY WORDS: scanner, calibration, color theory, polynomial transformation, least square adjustment, 3D transformation p
de
ABSTRACT: N
us
Digital imaging is a rapid growing area with the developments in computer technology. Digital color use has become a conventional SC
tool in many disciplines, such as cartography remote sensing and photogrammetry. The need for better color quality enforces digital TI
imaging industry to produce devices with less color distortions. Despite the today's advanced level of technology, it is known that 2
input and output devices cause color distortions, which depends on the quality of the device itself. Scanners, being peripheral devices ec
that capture the image of an object, are mostly used input devices in digital imaging. In this study color accuracy of desktop scanners SC
is handled. A method based on polynomial transformation, which makes possible to map device dependent colors to device ec
independents ones, is introduced. Tests on five different desktop scanners show the evident applicability of the method.
In
sa
1. INTRODUCTION 2. RESEARCH ON SCANNER CALIBRATION at
i : ; : ; : ; : 12
Digital color usage has been rapidly growing with the Baltsavias and Waegli (1996) found the geometric accuracies of
: : ie ^ ga
developments of software and hardware technology. Color desktop publishing. scanners used in photogrammetry and in
imaging has been involved with a variety of branches such as cartography between 4 and 7 micron. They performed the
printing industry, graphic arts, photography etc. The need for
better quality in color display and color output enforces the
digital imaging industry to develop better products. Scanners
have been used in many areas of digital imaging. These devices
capture the digital image of an object and save it as an image
file. The quality of the captured image depends on the scanner.
Since scanners can have lost their quality from time to time,
they have to be calibrated. The most simple and common way
to calibrate them is to compare the scanned image with the
original image. For this purpose certified calibration cards for
color accuracy and test targets for geometric accuracy have
been used. The colors and sizes of the certain objects on the
cards have been calibrated according to an international
standard. They can be used for color calibration or geometric
calibration. The cards are scanned, and then the color values
obtained by scanner are compared with the original color values
of the card. The calibration methodology is actually to
determine a transformation model between scanned values and
original values. Linear or non-linear models are possible. If
such a model is developed, every scanned image can be
corrected. One of the transformation models used in scanner
calibration is the polynomial transformation. In most
applications polynomial transformations have been determined
between the device dependent color space and one of the device
independent color spaces defined by CIE (Commission
Internationale de L'Eclairage). In this study, polynomial
transformations with different orders are determined based on
RGB (Red, Green, Blue) values of test cards and RGB values
obtained by scanner. For five different desktop scanners
polynomial transformations are compared with each other to
find the most suitable degree of polynomial. The approach here
can be applied for all kinds of scanners.
calibration in two steps. In the first step influences of lens
deformation was modeled. In the second step, using all points Ki
as control points an affine transformation was performed.
ap
: : va
Baltsavias (1996) concluded that the geometric accuracy of Cc
desktop publishing scanners was low. He suggested limiting x
cartographic scans to A3 size, checking and calibrating them Cal
regularly. ii
sta
Vrhel and Trussel (1999) present the mathematical formulation
of calibrating color scanners. They found that the mapping from Be
scanned values to colorimetric values is nonlinear. They applied :
S : re : me
artificial neural network for calibration, then compared this
method with other calibration methods based on a test a
performed with 264 samples. Ha
Hardeberg et al (1996) propose an analytic method based on 3rd
order polynomial regression techniques. They used CIE color i
space values and scanned values of 288 parts of the IT8.7/2
color calibration card. They found out that the polynomial Yi
regression delivers better results than other methods. pol
Finlayson and Drew (1997) mentioned that the color values
measured by color devices (e.g. scanners, color copiers, and m
color cameras) must be transformed to colorimetric e
*tristimulus" values in order to characterize them in a device
independent fashion. Two well-known methods for this
transformation are the simple least squares fit procedure and À à
Vrhel's principal component method. They propose a new obj
constrained regression method based on finding the mapping arr:
which maps a single (or possibly two) basis surface(s) without CC
error and also minimizes the sum of squared distances between ha
the mapped RGB data and corresponding XYZ tristimuli elec
890
