A JAVA-BASED SYSTEM FOR REMOTE CORRECTION OF CRT COLOR DISTORTION

A.Abrardo, M.Barni

Dept. Inform. Engin. University of Siena Siena, Italy
V.Cappellini, L.Fabiani, M.Zappalorti
Dept. Electron. Engin. University of Florence Florence, Italy
Abstract - Calibration of the output device used for the reproduction of digital colour images - a color CRT in most cases - can either be achieved through conventional techniques involving mathematical modeling of the CRT, or through a novel neural-network-based scheme, introduced in this work. A Java-based system for CRT remote calibration is also presented, which allows the user to get rid of the computational burden necessary to train the neural network, or to estimate the parameters of the CRT model by relying on a set of measurements of the colours displayed by the CRT. Measurements can be avoided as well, by storing calibration data relative to a wide variety of CRTs and using them to calibrate monitors with similar characteristics.

INTRODUCTION

An ever increasing number of applications call for high quality reproduction of digital colour images. Among them, remote access to digital pictures archives play an outstanding role in the eld of emerging remote multimedia applications. In this context we assume that the original scene was correctly digitized, i.e. pixel values correspond to true scene colours expressed in a standard, Device Independent (DVI), coordinate system (e.g. the CIE-RGB or CIE-XYZ systems 1, 2]). The most widely used reproduction devices for multimedia applications are the color Cathode Ray Tube (CRT) monitors. The behaviour of a CRT can be modeled as a multidimensional mapping from CRT control values to colours speci ed in a DVI space. When DVI colours have to be reproduced on a CRT, then, an inverse mapping transformation is needed to obtain the desired colorimetric output. The term calibration is often used to denote the entire procedure of characterizing a device and determining the inverse transformation. Usually a mathematical model is employed to describe the behaviour of the CRT, in such a case, calibration is readily achieved by determining the model parameters from a few measurements 1]. In many practical cases, though, the behaviour of the display is far from ideal and the mathematical
model does not t CRT's characteristics accurately enough. In such cases, a di erent approach is necessary to provide calibration. In this work we propose a new calibration procedure based on neural networks. A problem with neural-network-calibration is that network training may require a large amount of processing resources that may not be available at the remote user site. To get around the problem we propose a remote CRT calibration service, where a specialized centre carries out the colour calibration procedure based on few data provided by the remote user, i.e. some measurements of the characteristics of the remote colour output device. Besides, the use of standard calibration functions can be envisaged to deal with situations in which remote users are not able of carrying out the necessary measurements on their output device. CRT remote calibration should work across heterogeneous network environments, such as a corporate Intranet, or the global Internet. Java programs, being robust, secure, easy to use and to understand, and automatically downloadable from the Internet 3], are an excellent basis for this kind of applications.

CRT CALIBRATION

Reproduction of a light stimulus described by a triplet of coordinates in a DVI color space on a CRT screen involves two main steps: rst, DVI coordinates, here playing the role of CRT control values, are transformed into analog electrical signals, then the analog signals are used to stimulate CRT's phosphors. To describe mathematically the relation between the light emitted by phosphors and CRT control values, let Lpr max( ), Lpg max ( ) and Lpb max ( ) be the spectral radiance of CRT phosphors under the maximum input control values. The spectral radiance Lc( ) of any colour C reproduced on the monitor can be expressed as an additive mixture of Lpr max( ), Lpg max ( ) and Lpb max ( ), that is

) = Rp

Lpr max (

) + Gp

Lpg max (

) + Bp

Lpb max (
where Rp , Gp and Bp are the normalized coordinates of C in a color space whose primaries coincide with the lights emitted by the monitor phosphors. The description of CRT's behaviour consists in the determination of the relationship between Rp , Gp , Bp and CRT's input control values, say dr , dg , db. By referring to the analysis reported in the works by Berns et al. 4, 5] , it is derived for the red channel (kg r dr + ko r ) r (kg r dr + ko r ) 0 Rp = (2) 0 (kg r dr + ko r ) < 0 where kg r is the normalized gain of the red CRT channel, ko r is the normalized o set of the red channel, and r , is the so-called parameter
accounting for the non-linear behaviour of the CRT. Similar relationships may be given for the green and the blu channels, as shown in 4, 5]. To illustrate the calibration process, suppose we want to reproduce a color on a CRT monitor. Let us assume that pixel values are expressed in a DVI color system, e.g. the CIE-XYZ system. If the CRT transfer function (2) is known,and the linear transformation between the CIE-XYZ and the monitor phosphors systems is given, an inverse mapping function can be obtained, which for each desired output color expressed in the XYZ system gives the corresponding control values at the input of the CRT. The model parameters are calculated by applying the MSE criterion to a set of reference colors: the CRT is fed with a set of known control values and the corresponding XYZ output color coordinates are measured. The calibration procedure described above strictly relies on the mathematical characterisation of the CRT behaviour given in (2). If this model does not strictly t the actual CRT behaviour, the e ectiveness of model-based calibration is severely compromised and alternative calibration schemes must be used. The approach we propose to overcome the problems due to the non-ideal behavior of the CRT consists in approximating the relation between CRT control values and the colors reproduced on the screen through a neural network. As a matter of fact, we exploit here the well known capacity of neural networks to approximate any transfer function, which makes them a valid alternative to the use of precise models in cases where the system to be modeled is too complex to be described by means of mathematical formulae. As for the estimation of the parameters of the mathematical model, a set of reference colors is used to train the network. In the attempt to nd a reasonable trade-o between calibration accuracy and computational complexity (mainly residing in the training phase), extensive testing has been performed to set the architecture of the neural network. A rst simpli cation of the architecture, leading to considerable computation saving, consists in the use of three independent networks, one for each color channel, instead of a multiple output network. We have found experimentally that such a simpli cation does not preclude the accuracy of calibration, but at the same time it is much less computing demanding than a multiple-input multiple-output con guration. With regard to the architecture of each singular network, we used a strategy similar to that adopted by Schettini et al. 6] for color scanner calibration. As a consequence, and as a result of the tests we carried out, we found that a good compromise between accuracy and complexity is achieved by using a feed-forward, completely connected, 3-9-1 network trained by means of back-propagation. Experiments have been carried out to validate the calibration procedures described so far. The experiments regarded the calibration of a Sony Trinitron Display CPD-200GST controlled by a Pentium PC equipped with a Matrox Graphics MGA-G200 graphic board. Experiments relative to both modelbased and neural-based calibration have been performed by considering a

set of 73 test colors randomly spread over the whole color space. This set of colors have been split into two parts: the rst subset consists of 63 colors (training set) and it has been used to tune the parameters of the mathematical model and to train the neural network, whereas the second subset with the remaining 10 colors (test set) has been used to evaluate the e ectiveness of calibration. The XYZ coordinates of the light emitted by the CRT have been measured by using an Optic Multichannel Analyzer (OMA) produced by EG&G - Princeton Applied Research. To evaluate the impact of calibration errors on the visual color appearance of reproduced colors, the CIE-Lab di erence 2] between the target colors and those obtained by using the estimated control values has been measured. The average CIE-Lab di erence was ELab = 3:87 for the MSE approach, a value which indicates a good quality reproduction. In the case of neural network approach, the experimented average CIE-Lab di erence was ELab = 1:54, thus witnessing the superiority of the this approach to calibration.

REMOTE CRT CALIBRATION

The remote CRT calibration procedure has been developed through a three tiers client/server architectural model. This approach o ers many advantages with respect to traditional two tiers client/server model, thanks to middle tier introduction and to the separation between user interface management and data storage. By means of separation, data storage alteration has no impact on code that manages data visualisation and, at the same time, many independent representations of the same data are allowed. The di erent application layers have all been developed through Java language, which allows the use of the same executable code over di erent operating environments, be they home personal computers or departmental servers. Moreover Java o ers other signi cant features: Scalability, thanks to built-in multithreading support Security, thanks to virtual machine sandbox model Code migration, an Applet feature, which allows to use a program without previously installing it. A diagram of the three tiers remote CRT calibration procedure is reported in gure. A distributed components approach has been adopted for middleware implementation, according to the OMG (Object Management Group) CORBA (Common Object Request Broker Architecture) model 7], 8]. The ORB has been completely written in Java (the ObjectSpaces Voyager) that presents some unique features such as CORBA 2.0 compliance and exploitation of the Java platform characteristics (e.g. re ection and serialisation).

Applet Bean Bean Bean

TCP/IP
Mathematical Engine (Matlab)

ORB VOJAGER

Java Server Bean Bean Bean Bean Bean

Client 1

DBMS Server (Oracle)

Client N

Presentation tier

Middle tier

Back end tier
Figure 1: Block diagram of the remote three tier CRT calibration The presentation tier consists of a Java Applet which contains some Java components (JavaBeans) that attend to user interaction and information collection. Presentation tier components communicate with middle tier components thanks to ORB Voyager services. Middle tier components, which constitute the core of the application, consist of: Components that manage new calibration pro les generation, using data collected by user interface, and that apply these pro les to users images Components that store and retrieve calibration pro les to and from central database. The rst kind of components communicate with a mathematical engine (Matlab in this case), thanks to lower level components written in C++ language, that expose their interfaces through JNI (Java Native Interface). On the contrary, the second kind of components communicate with the DBMS (DataBase Management System) server through standard Java API JDBC
(Java DataBase Connectivity). Mathematical components check and prepare data that will be passed to the mathematical engine which, in turn, processes the data according to the calibration procedure described in the previous Section. Database access components manage all calibration pro les storing operations, like pro les insertion or deleting. Moreover, they attend to pro le searching procedure, which allows to retrieve previously calculated pro le according to simple searching criteria, like manufactor, model, CRT type, colour temperature, etc. Upon request of calibration, the server analyzes the information provided by the user and picks up from the archive the model which best ts the CRT to be calibrated. This functionality is envisaged to deal with situations in which remote users are not able of carrying out the necessary measurements on their output device.

CONCLUSIONS

We presented two methods to calibrate a CRT display. The rst one is based on a classical mathematical model of the CRT, whereas the second method is based on the approximation of CRT's behavior by means of a neural network. The results we obtained by testing the validity of both methods on a set of 10 reference colors showed that the neural approach o ers a great advantage in terms of calibration accuracy, the rationale for such an advantage being the ability of the neural network to cope with the possible non-ideal, non-linear behavior of the CRT. A remote Java-based CRT calibration service is also presented where a specialized centre carries out the colour calibration procedure based on few data provided by the remote user, i.e. the monitor type and its age. The proposed application can work across heterogeneous network environments, such as a corporate Intranet, or the global Internet.

ACKNOWLEDGMENTS

The Acknowledgment Section, if necessary, is inserted here.

References

1] G. Sharma, M. J. Vrhel, and H. J. Trussel, \Color imaging for multimedia," Proceedings of IEEE, vol. 86, no. 6, pp. 1088{1108, June 1998. 2] G. Wyszecki and W. S. Stiles, Color Science: Concepts and Methods, Quantitative Data and Formulae, 2nd edition, Wiley, New York, 1982. 3] A. van Ho , \The case for Java as a programming language," IEEE Internet Computing, vol. 1, no. 1, January-February 1997.
4] R. S. Berns, R. J. Motta, and M. E. Gorynski, \Crt colorimetry. part i: Theory and practice," Color Research and Applications, vol. 18, no. 5, October 1993. 5] R. S. Berns, R. J. Motta, and M. E. Gorynski, \Crt colorimetry. part ii: Metrology," Color Research and Applications, vol. 18, no. 5, October 1993. 6] R. Schettini, B. Barolo, and E. Boldrin, \Colorimetric calibration of color scanners by back-propagation," Pattern Recognition Letters, vol. 16, pp. 1051{1056, 1995. 7] R. Johnson T. Ulissides E. Gamma, R. Helm, Design Patterns: Elements of Reusable Object-Oriented Software, Addison-Wesley, 1995. 8] Steve Vinoski, \Corba: Integrating diverse applications within destributed heterogeneous environments," IEEE Communications Magazine., , no. 2, February 1997.

TRINITRON

CPD-100GST CPD-200GST
CB, ,. ,.7/2 : 966-01-01, 966-10-01
, . , , , . . , , .
, , . , , . , .
, 180 20 . , .

. :

240 30 . , 5. , , . .
1. . . 2. , . 3. GPE. ( . ). 3. MENU(). . 5. !/". " , . 6. /. . 7. . . , , .

8. . 9. (HD15). .

10. . 11. ( ).
, : (1) (1) Windows 95 (1) (1) (1) (1) IBM , VGA . , (CDP-100GST), 30 85(CDP200GST) Macintosh Power Macintosh , .

1: .

IBM PC/AT, Macintosh . , .

, , .

3: .
. : OUT OF SCAN RANGE NO INPUT SIGNAL . Windows 95. . ( ). VESA DDC Plug&Play. / DDC , Plug and Play Monitor (VESA DDC) Monitor Type ( ) Control Panel( ). / DDC. DDC, . , Monitor Type( ) (CDP-100GST CDP-200GST).

, . .

, , . AUDIO IN .

. 1. MENU.

CB, ,. ,.7/2 : 966-01-01, 966-10-01 .
2. !/" / SOUND, " MENU.

: .

3. !/" . "

30. MENU. RESET.

MUTING. .

MUTING.

( ). . . : , ( 50) .

, .

1. !/" /. "

BRIGHTNESS/CONTRAST (/).

/. .
CB, ,. ,.7/2 : 966-01-01, 966-10-01. .
!/". ".. . !.. .

3. RESET, . .

MENU . , .

1.COLOR

, .

2.CENTER

3.SCREEN

, .

4.ZOOM

5.OPTION

, , , .

6.SIZE

7.SOUND

.

8.GEOM

9.EXIT

CENTER.

. .

1. MENU.
2. !/" / CENTER " MENU.

!/". ". !.

/. .
30. MENU. RESET, . .

SIZE.

. .
2. !/" / SIZE " MENU.

, .

GEOM().

. , .

2. !/" / GEOM " MENU.

3. / , .

.

4. !/" . "

ROTATION ".. !.. PINCUSHION ". !. PIN BALANCE ". !. KEYSTONE ". !. . KEY BALANCE ". !. 30. MENU. RESET, .

ZOOM.

. .
2. !/" / ZOOM " MENU.

ZOOM.

3. !/" . "
".. !.. . 30. MENU. RESET, . : .

COLOR.

. , , , . .
2. !/" / COLOR " MENU.
(GPE) GPE , COLOR :
11000 9300. !/" . .

3. / .

. 9300 5000. 9300 5000. / VARIABLE !/".
".. ( ) !.. ( ). 30. MENU. RESET, . .

SCREEN.

, . CANCEL MOIRE , . .
2. !/" / SCREEN " MENU.

SCREEN.

CANCEL MOIR (ON) MOIRE ADJUST.
HI CONVERGENCE "... , !... , V CONVERGENCE ".. , !.. , CANCEL MOIRE ".. CANCEL MOIRE !.. CANCEL MOIRE MOIRE ADJUST ".. !.. 30. MENU. RESET, .

OPTION.

.

. .

2. !/" / OPTION " MENU.

OPTION.

3. / MANUAL DEGAUSS.

4. ".

, 20 . 30. MENU.

, (, , ).

3. / (OSD H POSITION) (OSD V POSITION).

(OSD H POSITION ) .

(OSD V POSITION )

3. !/" " .

30. MENU. RESET .

, POWER MENU.

3. / (CONTROL LOCK).
4. !/" ON. "
30. MENU. ON EXIT OPTION. , MENU POWER . . !/" OFF.

: , , , .

3. / (LANGUAGE).

4. !/" ".

ENG:, FRA:, DEU:, ESP:, JPN:. 30. MENU. ( ) RESET .
. 1. MENU, !/" / " , .

2. / , .

3. RESET.

. RESET.

. ( , ) .

. RESET .

, , .

(GPE).

GPE . GPE .

1. . 2. GPE .

GPE :
STANDARD , . PRESENTATION , . GRAPHICS/VIDEO , . 3. , , COLOR. : PRESENTATION GRAPHICS/VIDEO . , , . STANDART.
9 ( CDP-100GST) 10( CDP-200GST) plug&play. .
1.0 . >3.0 CDP-200GST. >3.6 CDP-100GST. >500 CDP-100GST CDP-200GST. Windows95. Windows , . CDP-100GST 1 VGA Graphics 640480 31.2 VESA 640480 43.3 VGA Text 720400 31.4 VESA 800600 46.5 VESA 800600 53.6 Macintosh 832624 49.16Color 7 VESA 1024768 60.8 VESA 1024768 68.9 VESA 12801024 64.CDP-200GST 1 VGA Graphics 640480 31.2 VESA 640480 43.3 VGA Text 720400 31.4 VESA 800600 46.5 VESA 800600 53.6 Macintosh 832624 49.16Color 7 VESA 1024768 60.8 Macintosh 1024768 60.19Color 9 VESA 10240768 68.VESA 80 75

, .

MENU .

30.

. : 1. <120(CDP-200GST) --------------------- <110(CDP-200GST) 2. < (1 ) 3. < (2 ) 4. < (3 ) 5. 0 ------------------------ . , NO INPUT SIGNAL. 30 3 . , .
. . Trinitron . , . , .

Plug&Play.

VESA DDC1 DDC2B ( ). DDC1 V.CLC VESA EDID( ). DDC2B .

, . 30.

: OUT OF SCAN RANGE , . NO INPUT SIGNAL , .
: . . NO INPUT SIGNAL : OUT OF SCAN RANGE , : 30-70(CDP-100GST),30-85(CDP-200GST) 50-120 . , . , , . , . , .
( ). , , , . . , 20.

, .

CB, ,. ,.7/2 : 966-01-01, 966-10-01 . . .
(. ). , , , . . , 20 , . . . , , . . . .

. . , . .

. Trinitron . , .

. . . .

. .
1. . 2. . 3. " 2.

(, , , ) , .

CB, ,. ,.7/2 : 966-01-01, 966-10-01 .

CPD-100GST

..0.25 , . ... 285214.....12801024 ... ...:30-70,:50-120....0.25() . , 80 .... /...100-240,50-60,1.8-1.0 .... 110...368392384.5(\\)..... 14

CPD-200GST

..0.25 , . ... 327243.....12801024 ... ...:30-85,:50-120....0.25() . , 80 .... /...100-240,50-60,1.9-1.1 .... 120...406432420(\\)..... 18

.....1

.....1.....1 .....1 ....1 ...1 ...2 .....2 .....2 .....3.....3 1: ...3 2: ....4 3: ...4 ....4 ...4 ....4 ....5 ....5 ....5 ....6 ....6 CENTER....7 SIZE....7 GEOM()....8 ZOOM....9 COLOR....10 SCREEN....11 OPTION....12 ....12 ...13 ....14 ....14 ....15 ....15 ...16 ...16 (GPE)..16 ...16 ....17 ....18 ....18 Plug&Play.....18 ...19 ....19 ....20 ....21