Three of the objects are made of identical gray paper while one is made of a different shade.

Post on 21-Jan-2016

24 views 0 download

Tags:

description

PERCEPTUAL STRATEGIES FOR MATERIAL IDENTIFICATION Qasim Zaidi Rocco Robilotto Byung-Geun Khang SUNY College of Optometry. Three of the objects are made of identical gray paper while one is made of a different shade. Illumination on the right is half the illumination on the left. 1. 2. 3. 4. - PowerPoint PPT Presentation

Transcript of Three of the objects are made of identical gray paper while one is made of a different shade.

PERCEPTUAL STRATEGIES FOR MATERIAL IDENTIFICATION

Qasim Zaidi Rocco Robilotto

Byung-Geun Khang

SUNY College of Optometry

1 2 3 4

•Three of the objects are made of identical gray paper while one is made of a different shade.•Illumination on the right is half the illumination on the left.

What is the number of the odd object?

3 1 2 4

“3” is the number of the odd object!

What strategy did you follow?

•Brightness discrimination threshold from % side-correct•Lightness identification threshold from % object-correct

Method of constant-stimuli

PREDICTIONS FOR TWO MODELS

Results of Lightness Identification

Hypothesis: P(I|D)=1.0Rejected at p=0.01Chi-squared > 16.8

Note systematic asymmetry in lightness identification compared to brightness discrimination.

1 2 3 4

•Choose the object that is most different in brightness.•Disregard the material and the illumination difference.

Brightness is different from photometer luminance because of adaptation and lateral interactions.

Results of Brightness Dissimilarity

Note systematic asymmetry in curves is similar to lightness identification results.

0

0.05

0.1

0.15

0.2

0.25

0.3

0 0.05 0.1 0.15 0.2 0.25 0.3

Exp 1

Ex

p 2

0

0.05

0.1

0.15

0.2

0.25

0.3

0 0.05 0.1 0.15 0.2 0.25 0.3

Exp 1

Ex

p 2

0

0.05

0.1

0.15

0.2

0.25

0.3

0 0.05 0.1 0.15 0.2 0.25 0.3

Lightness Identification (Exp 1)

Bri

gh

tne

ss I

de

nti

fica

tio

n (

Ex

p 2

)

0

0.05

0.1

0.15

0.2

0.25

0.3

0 0.05 0.1 0.15 0.2 0.25 0.3

Lightness Identification (Exp 1)

Bri

gh

tne

ss I

de

nti

fica

tio

n (

Ex

p 2

)

ST SS

ST SS

Brightness Discrimination Thresholds (Reflectance units)

Lightness Identification and Brightness Dissimilarity Thresholds (Reflectance units)

Half Illum.

Full Illum.

M ea n Lum in anc eo f S tim u lus

B r igh tn es so f S tim u lus

ga in

0

1

x

i

Adding adaptation to the photometer modelGain= k/(k+l)

0 x-0.5

0

0.5

1

x 3x-0.5

0

0.5

1

0.5x x

x 1.67x

0.83x

1.18x

(Fu

llIl

lum

.)(H

alf

Illu

m.)

No

rma

lize

dP

rop

ort

ion

of

Re

sp

on

se=

Tes

t

Test Reflectance

(a)

=

(b)

=0.5

(c)

=0.1

8

Lightness identification based on brightness dissimilarity responses of adaptation model

PERCEPTUAL STRATEGIES FOR LIGHTNESS IDENTIFICATION

•Under everyday conditions, observers consistently judge and remember surfaces as having a certain lightness or grayness.

•This phenomenology is not sufficient evidence that the visual system generically has access to the lightness of materials.

•For 3-D objects, we show conditions where lightness identification is limited solely by the limen of brightness discrimination, and other conditions where lightness identification is considerably worse.

•More importantly, we show that the same relative brightness based strategy reproduces both sets of results.

Differences between brightness and color information for material identification

Systematic changes in cone-coordinates

Algorithms for simultaneous identification for Lambertian and non-Lambertian 3-D objects

•Three of the filters are identical while one is different.•Illumination on the right is skylight, on the left is sunlight.

Which is the odd filter?

FILTER IDENTIFICATION ACROSS ILLUMINANTS

Identical filters will cause systematically different spectral changes in two spectrally different lights.

Given information about the filtered and unfiltered lights, can an observer identify two filters with identical spectral transmittances?

400 450 500 550 600 650 7000

1

2

3

4

5

6

7

8NATURAL LIGHTS (Taylor & Kerr,1941)

WAVELENGTH(nm)

RE

LA

TIV

E R

AD

IAN

T E

NE

RG

Y-- sunlight

-- skylight

400 450 500 550 600 650 7000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1COLOR FILTERS (Kodak CC50)

WAVELENGTH(nm)

TR

AN

SM

ITT

AN

CE

--- red --- green --- blue --- yellow --- magenta--- cyan

400 450 500 550 600 650 7000

1

2

3

4

5

6

7

8SUNLIGHT THROUGH SIX FILTERS

WAVELENGTH(nm)

RE

LA

TIV

E R

AD

IAN

T E

NE

RG

Y

--- red --- green --- blue --- yellow --- magenta--- cyan

400 450 500 550 600 650 7000

1

2

3

4

5

6

7

8SKYLIGHT THROUGH SIX FILTERS

WAVELENGTH(nm)

RE

LA

TIV

E R

AD

IAN

T E

NE

RG

Y

--- red --- green --- blue --- yellow --- magenta--- cyan

0.5 0.55 0.6 0.65 0.7 0.75 0.80

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

rg

b

y

m

cr

g

b

y

m

c

L/(L+M)

S/(

L+

M) R G B Y M C

0

0.02

0.04 SUNLIGHT

FILTERS

SD

(L/(

L+

M))

R G B Y M C0

0.02

0.04 SKYLIGHT

FILTERS

SD

(L/(

L+

M))

R G B Y M C0

0.02

0.04 SUNLIGHT

FILTERSS

D(S

/(L

+M

)) R G B Y M C

0

0.02

0.04 SKYLIGHT

FILTERS

SD

(S/(

L+

M))

R G B Y M C0

5

10

SUNLIGHT

FILTERS

SD

(L+

M+

S)

R G B Y M C0

5

10

SKYLIGHT

FILTERS

SD

(L+

M+

S)

Means of Material Chromaticities

SDs of Material Chromaticities

36 EVERYDAY MATERIALS (VRHEL et al 1994)

•Changes in means are systematic across illuminants.

•Filters can increase or decrease the color variation.

•Three of the filters are identical while one is different.•Illumination on the right is skylight, on the left is sunlight.

Which is the odd filter?

0 0.2 0.4 0.6 0.8 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1PSYCHOMETRIC CURVES

DELTA (CYAN VERSUS RED)

PR

OP

OR

TIO

N C

OR

RE

CT

SIDE CORRECT & FILTER CORRECT CURVES

•Method of constant stimuli•DELTA is distance between standard and distractor.•6 standards x 5 tests x 2 daylights x 6 deltas x 10 reps

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

BK

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

JM

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

JG

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

EK

IDENTIFICATION VS DISCRIMINATION

FILTERS

RESULTS

Identification thresholds were generally similar to discrimination thresholds and were significantly greater only in 5%, 10%, 13%, and 23% of 60 pairs for 4 observers.

Observers can identify filters across different illuminants almost as well as they can discriminate filters within the same illuminant.

MATERIAL IDENTIFICATION ACROSS ILLUMINANTS

X-junctions at the edges of the circular filters, in conjunction with the systematic changes in chromaticities of partially overlaid objects, promote a percept of transparency.

Are these results influenced by transparency. By rotating the image of the circular filtered regions by 180, we abolished the edge-based cues to transparency while leaving unchanged the figural and color contents inside and outside the filters.

•Three patches are from the same material while one is different.•Illumination on the right is skylight, on the left is sunlight.

Which is the odd material?

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

BK

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

JM

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

IDE

NT

IFIC

AT

ION

DISCRIMINATION

o Sunlight

+ Skylight

JG

IDENTIFICATION VS DISCRIMINATION

MATERIALS

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

T-J

UN

CT

ION

X-JUNCTION

o Sunlight

+ Skylight

BK

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

T-J

UN

CT

ION

X-JUNCTION

o Sunlight

+ Skylight

JM

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

T-J

UN

CT

ION

X-JUNCTION

o Sunlight

+ Skylight

JG

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

T-J

UN

CT

ION

X-JUNCTION

o Sunlight

+ Skylight

BK

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

T-J

UN

CT

ION

X-JUNCTION

o Sunlight

+ Skylight

JM

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

T-J

UN

CT

ION

X-JUNCTION

o Sunlight

+ Skylight

JG

IDENTIFICATION THRESHOLDS

DISCRIMINATION THRESHOLDS

FILTERS VS MATERIALS

RESULTS

Disrupting the X-junctions and color relations at the edges of the filters did not appreciably affect discrimination or identification.

Systematic color changes at the edges of the filter and X-junction cues to filter transparency were not necessary for good discrimination or identification performance.

COLOR IDENTIFICATION ACROSS ILLUMINANTS

•Both experiments can be conceived as identification of a circular ensemble of materials across the two natural lights.

•Both experiments indicate that identification of ensembles of materials is almost as good as discrimination between ensembles of materials.

•The significant cues are the chromaticity shifts across illuminants.

0.6 0.7 0.80.6

0.7

0.8RED

SK

YL

IGH

T

0.6 0.7 0.80.6

0.7

0.8GREEN

SK

YL

IGH

T

0.6 0.7 0.80.6

0.7

0.8BLUE

SUNLIGHT

SK

YL

IGH

T

0.6 0.7 0.80.6

0.7

0.8

YELLOW

SK

YL

IGH

T

0.6 0.7 0.80.6

0.7

0.8MAGENTA

SK

YL

IGH

T

0.6 0.7 0.80.6

0.7

0.8CYAN

SK

YL

IGH

T0.6 0.7 0.8

0.6

0.7

0.8NOFILTER

SK

YL

IGH

T

L/(L+M)

0 .03 .060

.03

.06

RED SK

YL

IGH

T

0 .03 .060

.03

.06

GREEN SK

YL

IGH

T

0 .03 .060

.03

.06

BLUE

SUNLIGHT

SK

YL

IGH

T

0 .03 .060

.03

.06

YELLOW SK

YL

IGH

T

0 .03 .060

.03

.06

MAGENTA SK

YL

IGH

T

0 .03 .060

.03

.06

CYAN SK

YL

IGH

T

0 .03 .060

.03

.06

NOFILTERSK

YL

IGH

T

S/(L+M)

0 20 400

20

40

RED SK

YL

IGH

T0 20 40

0

20

40

GREEN SK

YL

IGH

T

0 20 400

20

40

BLUE

SUNLIGHT

SK

YL

IGH

T

0 20 400

20

40

YELLOW SK

YL

IGH

T

0 20 400

20

40

MAGENTA SK

YL

IGH

T

0 20 400

20

40

CYAN SK

YL

IGH

T

0 20 400

20

40

NOFILTERSK

YL

IGH

T

L+M+S

CORRELATED CHROMATICITY CHANGES

•L/(L+M) coordinates translate

•S/(L+M) coordinates scale

•Means can be used to estimate translation and scaling constants

•Simple algorithms can exploit the affine transforms for material identification

# Observers Standard Test # Observer Standard Test

X- T- (Sunlight) (Skylight) X- T- (Skylight) (Sunlight)

________________________________________________________________________

_____

2 1 RED BLUE 3 2 BLUE MAGENTA

2 2 BLUE MAGENTA 1 1 MAGENTA RED

2 3 MAGENTA BLUE 2 2 BLUE RED

2 0 RED MAGENTA 1 1 RED BLUE

0 1 YELLOW GREEN 0 3 RED MAGENTA

3 1 CYAN BLUE 1 2 GREEN YELLOW

1 1 BLUE YELLOW

PAIRS OF CONFUSED FILTERS

Are RED and BLUE confused because observers judge deviations from directions of color vectors better than lengths of color vectors?

Means of Material Chromaticities

•If the material statistics are constant, then color identification only requires comparing rank-orders of chromaticities.•If material statistics change, affine transforms will give correct identification, but rank orders will not.

Color identification across illuminants within the same scene

PERCEPTUAL STRATEGIES FOR COLOR IDENTIFICATION

Identification based on similarity of appearance is affected by adaptation.

Identification based on systematic shifts of

appearance is based on geometrical operations in perceptual color space.

Identification strategies rely on associative

invariants of the natural world.