1. EE 638: Principles of Digital Color Imaging Systems
Lecture 14: Monitor Characterization and Calibration – Basic Concepts

2. Color Imaging Systems Goal:
Capture
Process
Output
Digital Camera
Scanners
Display: CRT
LCD
Projector
Printers: Laser EP IJ
Dye-sub
Liquid EP
offset
RGB
RGB
Device-dependent
CMYK
Goal:
want colors to look same through out the system.

3. Which CMYK? – effect of rendering device
HP DJ 970Cse
Mac Driver
HP DJ 970Cse
IPP Driver
Monitor

4. Which CMYK? – effect of capture device
Olympus C3000
Digital Camera
Heidelberg Scanner

5. Different color representations
Are they all equivalent?
How do we get from one to the other?
Can we get from one to the other?
Even if we can, what do all these numbers mean?

6. Example: capture to capture
Given RGB values from one capture device, can we predict RGB values for a second capture device?
Olympus C3000
Digital Camera
Scanner
RGB values
(23, 136, 180)
(203, 11, 52)
(219, 186, 33)
(7, 7, 7)
RGB values
(?, ?, ?)
(?, ?,?)

7. Two Approaches to Color Management
1. Closed pt-to-pt. solution
Separate mappings for each possible combination:
C1  P1
C1  P2
C2  P1
C2  P2
Camera 1
Camera 2
Printer 1
Printer 2

8. Two Approaches (cont.) Common Camera 1 Printer 1 Space CIE XYZ
2. Standard Interchange Space
Common
Space
CIE XYZ
Camera 1
Printer 1
Camera 2
Printer 2
Device
Dependent
Space
Device
Independent
Color Space
Device
Dependent
Space

9. Difficulty Increases CIE CRT RGB XYZ Task: Find mapping for
1) Display (CRT & LCD)
2) Capture (cameras & scanners)
3) printers
Once we have pieces we can use a color management system (CMS) to implement everything.
Development of transforms for CRT displays.
Goal: given XYZ, find RGB that produces that XYZ
Difficulty
Increases
CIE
XYZ
RGB
CRT

10. CRT Two steps: To do characterization need a device model DAC E-gun
1) characterize device
2) invert mapping (calibration)
To do characterization need a device model
DAC
E-gun
Digital Value
Shadow
Mask
CRT

11. CRT Magnified view of a shadow mask color CRT Magnified view of an
aperture grille color CRT

12. 3 Phosphor Types l P B
(l) G R
Primary Amounts

13. Overall (Forward) System Model
If primaries are visually independent, can find a 3x3 matrix , such that
Desired color
Necessary amount of primary
NL1
Input
Digital
Value
Displayed
CIE XYZ
NL2
NL3
Linear space of CRT monitor or LCD display

14. Single-Channel Excitation to Determine Nonlinearity
To get NLi , excite one channel at a time
Response for Y
Looking for (assume )
i.e. Digital
Values are
RGB
CRT
XYZ
Color Measurement Device
e.g. PR 705

15. Relation between Measured Y and Primary Amount + Multiplicative Scaling Assumption
(constant)
Note that this only works if changes to red channel model input multiplicatively scale the spectral power distribution

16. Offset-Gamma-Offset Model
Assumption is:
As I change Ri in monitor input
Output spectral distribution only changes by multiplicative constant
Typical model:

17. Monitor Characterization Process
To determine NLR , apply inputs for
NLR
NLG
NLB
Displayed
CIE XYZ
Input
Digital
Value
Linear space of CRT monitor or LCD display
CIE XYZ
Color Measurement Device
e.g. PR 705

18. Fitting Model to Data NL R
Measure corresponding Yi values model for nonlinearity:
“off”  “offset”
Once we know NLR, NLG, NLB can determine matrix T
Let
Repeat for G, B to entire matrix
NL R

19. Transformation between Linear RGB and CIE XYZ: Overdetermined Solution and Inclusion of Nonlinear Terms
To have more robust results, typically use a larger set input-output
Solve for T using least-squares
for over-determined systems.
Generalization of model :
See Osman Arslan paper
for example
Measured
Known
Question: Why do we need these nonlinear terms?

20. Evaluating Accuracy of the Model
How do we evaluate accuracy of calibration?
Have a box (monitor)
Completed characterization: NLR, NLG, NLB, T
Adjust
Physical
Device
Effective Device
Display

21. Evaluating Accuracy of the Model: Method 1
Examine how well model fits data based on data used to determine parameters (model-fit)
Examine how well model fits data based on data not used to determine parameters (cross-validation)
NLR
NLG
NLB
Forward Model
Proportional
to Photon Count
Gamma
Corrected
Space
Uncorrection
Outputs measured with a colorimeter or spectroradiometer
Outputs computed from the model
COMPARED WITH
Linear
Space

22. Evaluating Accuracy of the Model: Method 2
Calibration
Process
Physical
Device
PR705
Method 3a: Compare
for some set of colors of
interest, and compute
Method 3b:Use human viewer to do psychophysical assessment
Pros: Accounts for entire system quantization (noise,
instability …) or bottom line
Cons: Requires measurements in lab, i.e. time and effort

23. Some local color history
John Dalton
MSEE University of Delaware, circa 1983
Worked for Textronix, Wilsonville, OR on inkjet printers with Chuck Johnson (Zhen He worked there, now at Intel)
Worked for Apple with Gary Starkweather (inventor of laser printer, now at Microsoft)
Founded Synthetik and moved to Hawaii
Chuck Johnson
Left Textronix to join start-up Mead Imaging, Dayton, OH
Contacted me to do research on color in 1985
Ron Gentile
Interned at Mead Imaging
Ph.D. Purdue, 1989
Early employee at Adobe
Co-founded Bellamax

25. text

26. Experimental results for gray balancing (NLi) (Gentile et al, 1990)

27. Experimental results for forward model (Gentile et al, 1990)

28. Additional resource for display device characterization and calibration
Minh_Nguyen_Monitor_Calibration.pptx (can be found in Reference section of course website)
Features
Summary and review of work by Arslan, Thanh, and Min
Detailed discussion of how to set white point
Description of three different models for gray balance curve
Gamma-based
Two part gamma-based
Spline curve
Recent experimental results
Achieves 4 Delta E average error with gamma-based
Less than 2 Delta E average error with other two methods listed above
Documents day-to-day variability