An Efficient Uniform Color Space for High

Dynamic Range and Wide Gamut Imagery

1

June 27, 2017

Presenter: Muhammad Safdar (PhD)

Supervisor: Prof. M. Ronnier Luo

Zhejiang University, China

ICC-CMRF (2016-2017)

Uniform Color Space Predicts perceptual color difference

No inter-dependence b/w lightness, chroma, and hue

2 • K. A. G. Smet et al., LEUKOS, 2016 • I. Lissner et al., IEEE Trans. Image Process., 2012

Hue

Lig

htn

ess

2 2 2( ) ( ) ( )E L C H

Criteria for a New UCS

To uniformly encode high dynamic range signals e.g., 0-10,000 cd/m2

To uniformly encode wide gamut image signals e.g., Rec.2020

Uniformity (Local & Global) Hue Linearity Grey-scale Convergence

Chroma

Lig

htn

ess

3

1 2 3

4

5

Test Spaces

1) CIELAB

2) CAM16-UCS

3) ICTCP

4) Jzazbz

CIE/ISO standard

Accurately predicts small color difference data

Dolby’s proposal for HDR and WCG

Current Proposal

4 • C. Li et al., Color Imaging Conf., 2016 • Dolby, 2016 • CIE, Colorimetry, 2004

Criteria and Visual Data No. Criteria Data sets Purpose

1 Perceptual Color Difference

i) COMBVD ii) OSA

i) Training ii) Testing

2 Perceptual Uniformity i) COMBVD Ellipses ii) MacAdam Ellipses iii) Munsell Data

i) Reference ii) Testing iii) Testing

3 Hue Linearity i) Hung & Berns ii) Ebner & Fairchild iii) Xiao et al.

i) Reference ii) Testing iii) Testing

4 Wide-range Lightness Prediction

i) RIT SL1 ii) RIT SL2

i) Testing ii) Training

5 Grey-scale convergence

Chroma-ratio metric (%)

5 • D. L. A. MacAdam, JOSA, 1942 • D. L. A. MacAdam, JOSA, 1974 • M. D. Fairchild, Electronic Imaging, 2011

3Chroma-Ratio 100 w

r g b

C

C C C

Proposed Jzazbz space

6

D65

D65

D65

0.41478972 0.579999 0.0146480

-0.2015100 1.120649 0.0531008

-0.0166008 0.264800 0.6684799

L X

M Y

S Z

1 2

' ' '

3

, ,

10,000, , ,

, ,1

10,000

pn

n

L M Sc c

L M SL M S

c

'

'

'

0.5 0.5 0

3.524000 -4.066708 0.542708

0.199076 1.096799 -1.295875

z

z

I L

a M

b S

1

1

z

z

z

d IJ

dI

(1)

(2)

(3)

(4)

(5)

14where 2610 / 2n

where 0.56d

51.7 2523 / 2p

'D65 D65D65

'D65 D65D65

( 1)

( 1)

bX b ZX

gY g XY

-50 0 50 100

-50

0

50

100

CIELAB

a*

b*

-40 -20 0 20 40-40

-30

-20

-10

0

10

20

30

40

CAM16-UCS

aM

bM

-0.2 -0.1 0 0.1 0.2-0.2

-0.1

0

0.1

0.2

ICT

CP

CP

-CT

-0.1 -0.05 0 0.05 0.1 0.15-0.1

-0.05

0

0.05

0.1

0.15

Jza

zb

z

az

bz

Results (Uniformity)

7

-CT bz

(a) (b)

CPaz

ICTCP Jzazbz

8

Results (Uniformity in WCG)

Results (Quantitative)

9

0

10

20

30

40

50

60

70

80

COMBVD OSA COMBVD

Local

COMBVD

Global

MacAdam

Local

MacAdam

Global

CIELAB

CAM16-UCS

ICTCP

Jzazbz

CIELAB

CAM16-UCS

ICTCP

Jzazbz

10

Results (Hue Linearity)

Results (Quantitative)

11

0

2

4

6

8

10

12

14

Hung & Berns Hung & Berns

(blue)

Ebner & Fairchild Xiao et al.

CIELAB

CAM16-UCS

ICTCP

Jzazbz

CIELAB

CAM16-UCS

ICTCP

Jzazbz

Results (Lightness Prediction)

12

Results (Lightness)

13

0

5

10

15

20

25

30

35

RIT SL1 RIT SL2 Munsell Value Chroma Ratio

(%)

CIELAB

CAM16-UCS

ICTCP

Jzazbz

CIELAB

CAM16-UCS

ICTCP

Jzazbz

Jzazbz

Second best for small color difference

Best for large color difference

Best for wide gamut uniformity

Best for hue linearity

Best for wide-range lightness prediction

Plausible grey-scale convergence

Jzazbz gave overall best performance and should be confidently used

Conclusions

14

Demonstration (Image Segmentation)

Original CIELAB CAM16-UCS

YCbCr ICTCP Jzazbz

Segmentation (7 regions) using K-means clustering algorithm

1. Journal Paper: Optics Express (IF=3.3)

https://www.osapublishing.org/oe/abstract.cfm?uri=oe-25-13-15131

2. MATLAB Code for Jzazbz

Deliverables

15

Any

Questions?

Model Development Initially, we start using a structure similar to ICTCP

1,1 1,2 1,1 1,2 D65

2,1 2,2 2,1 2,2 D65

3,1 3,2 3,1 3,2 D65

1

1

1

L X

M Y

S Z

'

1,1 1,2 1,1 1,2

'

2,1 2,2 2,1 2,2

'

3,1 3,2 3,1 3,2

1

1

1

z

z

I L

a M

b S

1 2

' ' '

3

, ,

10,000, , ,

, ,1

10,000

pn

n

L M Sc c

L M SL M S

c

(1)

(2)

(3)

• Dolby, 20016 • M. Safdar et al., Color Imaging Conf., 20016

Then we introduced a linear given below to correct hue linearity

'D65 D65D65

'D65 D65D65

( 1)

( 1)

bX b ZX

gY g XY

(1)

'

1,1 1,2 1,1 1,2 D65

'

2,1 2,2 2,1 2,2 D65

3,1 3,2 3,1 3,2 D65

1

1

1

L X

M Y

S Z

'

1,1 1,2 1,1 1,2

'

2,1 2,2 2,1 2,2

'

3,1 3,2 3,1 3,2

1

1

1

z

z

z

I L

a M

b S

1 2

' ' '

3

, ,

10,000, , ,

, ,1

10,000

pn

n

L M Sc c

L M SL M S

c

(2)

(3)

(4)

Model Development

• G. Kuehni., Col. Res. Appl., 1999 • G. Cui et al., Col. Res. Appl., 2002

Another simple equation to tune perceptual lightness

1

1

z

z

z

d IJ

dI

after Eq.5

(5)

Model Development

• M. R. Luo et al., Col. Res. Appl., 2006