Towards robust categorical colour perception
3 likes885 views
The document discusses a study on improving categorical color perception through crowd-sourcing methods. It outlines problems with current color categorization, the distinction between global and local color differences, and proposes a framework for collecting diverse color names in various languages. The ongoing research aims to develop a robust color thesaurus while addressing challenges such as variability in naming and participant feedback.
![Development of colour naming
Colour naming is acquired, not genetic
socio-economic status (SES)
Franklin et al., PNAS 105(9): 3221–3225, 2008
900 550
infants within-category adults
initiation time [ms]
initiation time [ms]
between-category
800
within-category 450
between-category
700
350
600
500 250
left right left right
visual field visual field
Occurs late in child’s development, but age is decreasing with
increase of technology
1900: basic four colours @ 8 years
1950: @ 5 years of age
Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 16 / 31](https://image.slidesharecdn.com/gbbaic09-090923142912-phpapp02/85/Towards-robust-categorical-colour-perception-16-320.jpg)
Towards robust categorical colour perception
- 1. Towards robust categorical colour perception G. Beretta N. Moroney J. Recker Print Production Automation Lab Hewlett-Packard Laboratories Palo Alto, California 11th Congress of the International Colour Association Sydney, 27 September – 2 October 2009 Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 1 / 31
- 2. Outline 1 Problems 2 Global vs. local colour differences 3 What and where are the categories? 4 New paradigm: use crowd-sourcing 5 Status & conclusions Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 2 / 31
- 3. Outline 1 Problems 2 Global vs. local colour differences 3 What and where are the categories? 4 New paradigm: use crowd-sourcing 5 Status & conclusions Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 3 / 31
- 4. Describing colours Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 4 / 31
- 5. HTML colour specification #EF4123 #007CB0 #848688 #BF1E74 #007CB0 #008F4C #F89F6D #F499B8 Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 5 / 31
- 6. Automatic layout in variable data printing HP chiclet HP chiclet old palette new palette automatic layout Buy HP Buy HP computers workstation xw 8600 Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 6 / 31
- 7. The wide xvYCC gamut Luma Gamut of xvYCC Y 254 Over White 1.0 235 1< R’,G’,B’ 1< R’,G’,B’ BT.709-5 (sRGB) sYCC Extended Region Extended Region 0 < R’,G’,B’ < 1 xvYCC (Gamut of BT.709-5) (sRGB) R’,G’,B’< 0 R’,G’,B’< 0 0.0 16 -0.57 - 0.5 Black +0.5 +0.56 1 128 Cb, Cr 1 16 240 254 Extended Extended Chroma Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 7 / 31
- 8. Applications of colour naming Better user experience in GUIs Automatic nudging of text and logo colours for readability in variable data printing Gamut mapping for HDR and wide gamut displays Culture-independent preferred color rendering Thematic rendering Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 8 / 31
- 9. Outline 1 Problems 2 Global vs. local colour differences 3 What and where are the categories? 4 New paradigm: use crowd-sourcing 5 Status & conclusions Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 9 / 31
- 10. Local colour differences y 520 530 0.8 540 510 550 Stiles Line Element Ellipses plotted 3 x 560 0.6 570 500 580 590 0.4 600 610 620 490 630 700 0.2 480 470 0 460 x 45 0 0.2 0.4 0.6 Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 10 / 31
- 11. Categorisation Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 11 / 31
- 12. Categorical perception Definition (Stevan Harnad) A categorical perception effect occurs when 1 a set of stimuli ranging along a physical continuum is given one label on one side of a category boundary and another label on the other side and 2 the subject can discriminate smaller physical differences between pairs of stimuli that straddle boundary than between pairs that are entirely within one category or the other v’ v’ 1. 1. 1. white yellow white yellow .98 .92 .65 .53 orange 1. 1. 1. .73 87 .91.87 .55 green1. .99 .98 .71.53 .75.88 orange .97 .88 .62 .96.98 .98.96 .98 .92 .65 .53 .59.73 .74.57 .93 .97 .91 .94 .33 .45.51 .84 .66 .5 .46.44 A .58.54 .63.59 .73 87 .91 .31.49.63 .52 .5 green 1. .99 .98 .97 .88 .71 .71 .68 .78.81 .63.47 .7 .44 .47.52 .33.39.61 red .55 .87 .71.53 .41.35.56 .57.56 .44 .61.61.52 B .33 .45.56 .56 .53.53 .47.47 .75.88 .96 .43.32.46 aqua .77.63.38.39.56C E D .64.53.35 .56.68 .76 .74.75.58 peach .62 .59.73 .98 .98.96 .45 .53 .7 .8 .82.65.36.45 .85.87.82.74 .45.48 .3 .74.57.33 .52.74.81.87 .93 .69.42 .36 .91.89 .86 .47.48.28 .97 .91 .94 .45.51 .58 .33.57.69.83 A .53.64 .5 .92.88.87 .38.37.48.44 gray .62.75 .84 .66 .5 .46 .54 .63.59.52 .57 .7 .82.86 .53.35.56.56 .47 .56.68 .88 pink .44 .31.49 .72.74 .5 .74.82.53 .54.67 .77.59 .48 .35 .51.63 .63 .78 .81 .63 .9 .84 .48 .71 .71 .68 red .7 .83 .8 .72.65 .5 .92.89.53 .7 .44 .47 .47 .33.39.61 .82.92 .9 .75.69 .52.41.35 .98 .9 .52 .83.93 .89.81 purple blue .98 .9 .53 .83.94.92 .57.56 .44 .56.61.61 B .52 .53.53 .97 .9 .52.91 .95 .25 .33 .45.56 .56 .96.91 .6 .84 A = CIE Standard Illuminant A .97.92.63 .43.32.46 .47.47 B = CIE Standard Illuminant B aqua .77.63.38.39.56C E C = CIE Standard Illuminant C D .56 .68 .76 .97.94 D = CIE Standard Illuminant D65 E = equal-energy point .74.75.58 peach .97 u’ .15 .64.53.35 .53 .7 .8 .15 .35 .45 .45 .05 .25 .82 .65.36.45 .85 .87.82.74 David L. Post, 1988 .45.48 .3 .52.74.81 .87 .91.89 .69.42 .36 robust categorical colour perception Beretta, Moroney, Recker (HP Labs) .47 Towards .48.28 .86 AIC 2009 12 / 31
- 13. Global colour differences — how to find them? broken warm white A = 20 cement grey Roman ochre Pompeian yellow 100 90 80 70 60 50 Indian orange V 40 orange ochre 30 Arsigont 20 brown beige 10 Anatolian brown 0 0 10 20 30 40 50 60 70 80 90 100 T Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 13 / 31
- 14. Outline 1 Problems 2 Global vs. local colour differences 3 What and where are the categories? 4 New paradigm: use crowd-sourcing 5 Status & conclusions Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 14 / 31
- 15. Colour ontogeny of languages Brent Berlin and Paul Kay, University of Berkeley, 1969 The physiology underlying even the unique hues is unknown There is no natural categorisation orange and/or green yellow white pink and red blue brown and/or black purple and/or yellow green gray I II III IV V VI VII Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 15 / 31
- 16. Development of colour naming Colour naming is acquired, not genetic socio-economic status (SES) Franklin et al., PNAS 105(9): 3221–3225, 2008 900 550 infants within-category adults initiation time [ms] initiation time [ms] between-category 800 within-category 450 between-category 700 350 600 500 250 left right left right visual field visual field Occurs late in child’s development, but age is decreasing with increase of technology 1900: basic four colours @ 8 years 1950: @ 5 years of age Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 16 / 31
- 17. Outline 1 Problems 2 Global vs. local colour differences 3 What and where are the categories? 4 New paradigm: use crowd-sourcing 5 Status & conclusions Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 17 / 31
- 18. Goal 1 Large dictionary currently harvesting unconstrained names extensive, through crowd-sourcing evolves through time not limited to one language 2 Number of synonym categories 12 decided though crowd-sourcing not 266 like in ISCC–NBS thesaurus . . . or 26, or 30, or 80. . . 3 Algorithm for determining categories construct separate categorisations for each colour patch explicitly ask user for a specific and a general name explore boundary-finding algorithms Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 18 / 31
- 19. Multilingual colour naming experiment http://www.hpl.hp.com/personal/Nathan_Moroney/mlcn.html Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 19 / 31
- 20. The colour thesaurus http://www.hpl.hp.com/personal/Nathan_Moroney/color-thesaurus.html Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 20 / 31
- 21. From colour naming to thesaurus conventional exclusionary usage corpus naming typographic raw corpus deletions substitutions spell-checker experiment harmonization inverse YCiCii for each substring antonyms merging neighbors name statistics CIECAM02 synonyms neighbors threshold core frequency scrubbed number unique vocabulary analysis corpus IP addresses Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 21 / 31
- 22. Contributed name distribution 2000 green blue light blue lavender grass green tan khaki dark brown cornflower cream purple gray yellow green dark purple red orange pink navy blue navy moss green dark red red maroon peach burnt orange chocolate 1600 black lime green lime dark blue burgundy salmon spring green pea green crimson coral brown dark green light purple baby blue apple green magenta lilac rose kelly green eggplant violet olive gold dark pink goldenrod sky blue olive green plum rust medium blue 1200 orange yellow cyan periwinkle brick red beige blue green fluorescent green ocean blue leaf green teal mint green mustard sage bright purple light green bright green white hunter green grape fuchsia mauve indigo pale green light yellow 800 turquoise aqua sea green hot pink bright blue chartreuse blue gray cobalt emerald jade royal blue neon green light brown midnight blue ochre forest seafoam aquamarine light pink army green brick 400 0 gre r m y t l m l p s g p r b b k b k fl b c c e l e b en ed agen ellow urquo ight b aroo ilac eriwi ea gr rass g each ose eige right haki urnt elly g uores lue g ornflo hocol ggpla eaf gr mera rick ta ise lue n nkl een re blu ora ree ce ray we ate nt ee ld e en e nge n nt g r n ree n Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 22 / 31
- 23. Handling missing names naming corpus raw corpus experiment scrubbing name lexical harvesting analysis FALSE crowd name synonyms and TRUE found antonyms Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 23 / 31
- 24. Expanding the corpus Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 24 / 31
- 25. Robustness: qualifying the corpus Problems: we observed about 3% disruptive participants in the experiment variability of rarely used names Solution is to collect explicit feedback on the global statistics from each participant More efficient than recruiting domain specialists Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 25 / 31
- 26. Feedback distribution correct, spot on good neutral poor wrong, completely wrong Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 26 / 31
- 27. Outline 1 Problems 2 Global vs. local colour differences 3 What and where are the categories? 4 New paradigm: use crowd-sourcing 5 Status & conclusions Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 27 / 31
- 28. What we have so far Framework for collecting colour names in multiple languages Robust: colour thesaurus served 194’369 color names as of Wednesday the 23rd of September 2009 Robust: feedback mechanism for improving the corpus quality with use Mechanism to harvest less common names Still cheating on categorisation: synthetic synonyms vs. categories, but making progress . . . Experimenting with linguistic analysis tools to find category boundaries No user interface yet to collect category names and antonyms Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 28 / 31
- 29. Is the category hierarchy just 2 deep? broken warm white cement grey Anatolian brown Roman ochre yellowish orange 1 brown beige yellowish orange 2 Arsigont yellow orange 1 Pompeian yellow orange orange 2 orange ochre Indian ochre red orange 3 violet reddish orange 1 reddish orange 2 blue green1 green2 Coloroid color 369 names + 7 domains 48 basics 79 synonyms Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 29 / 31
- 30. Height 0 1 2 3 4 shiny plastic gloss sticky medium.weight medium photo waxy ultra flat coated vinyl smooth Beretta, Moroney, Recker (HP Labs) dull diffuse off.white semi semi.gloss transparent clear thin light.weight Albuquerque, November 2009 pale ivory yellow cream beige soft eggshell chalky blue green gray satin decorative grain texture rough brown pearl Towards robust categorical colour perception metallic silver high parchment Dendrogram — good things to come . . . tan office matte paper white art rigid surface color canvas linen fine pattern heavy.weight AIC 2009 heavy card thick stiff see: Moroney & Beretta,“Nominal scaling of print substrates,” CIC 17, bright gold 30 / 31
- 31. Questions and Discussion http://www.hpl.hp.com/personal/Giordano_Beretta/ http://www.hpl.hp.com/personal/Nathan_Moroney/ http://www.hpl.hp.com/people/john_recker/ blog: http://mostlycolor.ch Beretta, Moroney, Recker (HP Labs) Towards robust categorical colour perception AIC 2009 31 / 31