I2
DICOM Grayscale Standard Display
Function
David Clunie
Quintiles Intelligent Imaging
Clear Vision for the Healthcare Industry
Outline
• Inconsistent appearance of images
– Why is it a problem ?
– What are the causes ?
• Grayscale Standard Display Function
– The DICOM solution to the problem
– How it works
– How to implement it
Distributed Image Consistency
Laser Printer
Digital Modality
Identical perceived contrast
Workstation
Workstation
Distributed Image Consistency
Laser Printer
Digital Modality
Identical perceived contrast
Workstation
Workstation
Distributed Image Consistency
Laser Printer
Digital Modality
Identical perceived contrast
Workstation
Workstation
Distributed Image Consistency
Laser Printer
Digital Modality Identical perceived contrast
and color !!
Workstation
Workstation
What about color ?
• Consistency is less of an issue:
– US/NM/PET pseudo-color; VL true color ??
• Consistency is harder to achieve
– Not just colorimetry (i.e. not just CIELAB)
– Scene color vs. input color vs. output color
– Gamut of devices much more variable
– Greater influence of psychovisual effects
• Extensive standards efforts e.g. ICC
Problems of Inconsistency
• VOI (window center/width) chosen on one
device but appears different on another
device
• Not all gray levels are rendered or are
perceivable
• Displayed images look different from
printed images
• …
Problems of Inconsistency
•VOI chosen on one
display device
•Rendered on another
with different display
•Mass expected to be
seen is no longer seen
mass visible mass invisible
Problems of Inconsistency
0.5 1.0
•Not all display levels
are perceivable on all
devices
1.5 3.0
Problems of Inconsistency
0.5 1.0
•Not all display levels
are perceivable on all
devices
1.5 3.0
Problems of Inconsistency
•Printed images don’t look
Digital Modality like displayed images Laser Printer
Causes of Inconsistency
• Gamut of device
– Minimum/maximum luminance/density
• Characteristic curve
– Mapping digital input to luminance/density
– Shape
– Linearity
• Ambient light or illumination
Causes of Inconsistency
•Display devices
vary in the maximum
luminance they can
produce
•Display CRT vs. film
on a light box is an
extreme example
1.0 .66
Monitor Characteristic Curves
Monitor Characteristic Curve
100 Gamma Maximum
Luminance
10
Ambient Light
0.1
0 50 100 150 200 250 300
Digital Driving Level
Towards a Standard Display
• Can’t use absolute luminance since
display capabilities different
• Can’t use relative luminance since shape
of characteristic curves vary
• Solution: exploit known characteristics of
the contrast sensitivity of human visual
system - contrast perception is different
at different levels of luminance
Human Visual System
• Model contrast sensitivity
– assume a target similar to image features
– confirm model with measurements
– Barten’s model
• Grayscale Standard Display Function:
– Input: Just Noticeable Differences (JNDs)
– Output: absolute luminance
Standard Display Function
Grayscale Standard Display Function
4500
4000
3500
3000
2500
2000
1500
1000
500
0
0 200 400 600 800 1000 1200
JND Index
Standard Display Function
Grayscale Standard Display Function
4500
4000
3500
Monitors Film
3000
2500
2000
1500
1000
500
0
0 200 400 600 800 1000 1200
JND Index
Standard Display Function
Grayscale Standard Display Function
1000
100
10
1
0 200 400 600 800 1000
.1
.01
JND Index
Standard Display Function
Grayscale Standard Display Function
1000
Film
100
10
Monitors
1
0 200 400 600 800 1000
.1
.01
JND Index
Perceptual Linearization
• JND index is “perceptually linearized”:
– same change in input is perceived by the
human observer as the same change in
contrast
• Is only a means to achieve device
independence
• Does not magically produce a “better”
image
Perceptual Linearization
Grayscale Standard Display Function
1000
Despite different change
in absolute luminance 100
10
1
0 200 400 600 800 1000
.1
.01
JND Index
Same number of Just Noticeable Difference == Same perceived contrast
Perceptual Linearization
Ambient Light
Display
Display Perception of Contrast
By Human Visual System
Modality
Using the Standard Function
• Maps JNDs to absolute luminance
• Determine range of display
– minimum to maximum luminance
– minimum to maximum JND
• Linearly map:
– minimum input value to minimum JND
– maximum input value to maximum JND
– input values are then called “P-Values”
Monitor Characteristic Curve
Monitor Characteristic Curve
100
10
Ambient Light
0.1
0 50 100 150 200 250 300
Digital Driving Level
Standard Display Function
Grayscale Standard Display Function
1000
Maximum Luminance 100
+ Ambient Light
Monitor’s Capability
10
1
0 200 400 600 800 1000
Minimum Luminance
+ Ambient Light
.1 Jmax == P-Value of 2n-1
.01
Jmin == P-Value of 0
JND Index
Standardizing a Display
100
Standard
10
1
0 50 100 150 200 250
Characteristic Curve
0.1
DDL or P-Values
Standardizing a Display
Mapping P-Values to Input of Characteristic Curve (DDL’s)
300
250
200
DDL
150
100
50
0
0 50 100 150 200 250 300
P-Values
Standardizing a Display
Standard Display Function
Standardized
Display
P-Values: 0 to 2n-1
Device Independent Contrast
Standard Display Function Standard Display Function
Standardized Standardized
Display A Display B
P-Values: 0 to 2n-1
So what ?
• Device independent presentation of
contrast can be achieved using the
DICOM Grayscale Standard Display
Function to standardize display and
print systems
• Therefore images can be made to
appear the same (or very similar) on
different devices
So what ?
• Images can be made to appear not only
similar, but the way they were intended
to appear, if images and VOI are
targeted to a P-value output space
• New DICOM objects defined in P-values
• Old DICOM objects and print use new
services (Presentation State and LUT)
Not so hard …
• If you calibrate displays or printers at all,
you can include the standard function
• If you use any LUT at all, you can make
it model the display function
• If you ignore calibration and LUTs totally
(e.g. use window system defaults) the
results will be inconsistent, mediocre and
won’t use the full display range