Is Shininess A Basic Feature In Visual Search? F111 · Is Shininess A Basic Feature In Visual...

1
Is Shininess A Basic Feature In Visual Search? Randall S. Birnkrant Jeremy M. Wolfe Melina A. Kunar Matthew Sng Brigham & Women's Hospital Brigham & Women's Hospital & Harvard Medical School Research Science Institute student A limited set of attributes (e.g. color, size, and a variety of depth cues) can guide the deployment of attention in visual search. Do monocular cues to shininess serve as guiding attributes? INTRODUCTION Rensink & Cavanagh (1994) found rapid search for a cylinder with a top highlight, suggesting information about specular reflectance is available at early levels of vision. CONCLUSIONS General Method Observers (Os) were told to respond quickly and accurately. They were also instructed to minimize eye movements. Stimuli remained on-screen until a response was made. Targets were present in approximately 50% of trials. Searches were blocked by target. Does shininess guide attention? Static Stimuli Experiment 2 Experiment 3 Experiment 5 Subjective ratings of shininess by a separate group of participants (n=11). Experiment 4 Stimuli: Shiny spheres rendered using photographically captured real-world illuminations were taken from the work of Fleming, Dror, & Adelson (2003). Matte spheres were created by carving up the specular reflections and replotting them. Clearly, the latter distorts global shape information. However, this is only a potential confound if these searches proved to be efficient. Results : Both searches proved inefficient. Shiny target #1 did produce a noticeably shallow search slope (7 ms/item, shininess rating: 8) and deserves further investigation. Stimuli: Os searched for targets defined by conjunctions of shape (spherical or rectangular) and shininess. Thus, a search target could either be a plain sphere, a shiny sphere, or a rectangle cropped from the same spherical specular reflection. If shininess is a feature, shininess-shape conjunctions should yield efficient searches. F111 Results : SEARCH FOR THE CONJUNCTION OF SHININESS AND SHAPE WAS NOT EFFICIENT. Its not that these stimuli are too complex or noisy to support efficient search. Shiny rectangles among shiny and matte spheres popped out, as did matte spheres among two types of shiny objects. However, observers could not use the combination of shininess and shape to guide attention. Stimuli: Similar to Experiment 1, shiny and matte spheres were distinguished by either a white highlight or dark patch. The spheres were now textured (rotating uniform spheres are pretty dull) and oscil- lated through an angle of ~ 70 deg at 3 Hz. Again, disks containing the opposite dot polarity were placed just above the spheres to thwart search strategies dependent on locating a particular dot polarity in the display. Control conditions were run in which Os searched for a specific configuration of light and dark patches on a flat surface. Results: Although target present RT x Set size slopes were quite shallow, SEARCHES WERE NOT FEATURE- LIKE. Again, it does not appear that Os were searching for a particular configuration of light and dark patches. Stimuli: Hartung & Kersten (2002) created compelling shiny and matte “kettles” where surface appearance was defined by the motion of specular reflections. Shiny kettles had dynamic specular reflections, whereas matte kettles had a “sticky” reflection - a single instance of specular reflec- tion that rotated with the kettle. Stimuli oscillated through an angle of ~ 30 deg at 0.6 Hz. Rotational phase was ran- domly chosen for each stimulus to avoid grouping effects. The kettles did oscillate about different axes, but this is only a potential confound if the searches proved to be effi- cient. 1. Although monocular information regarding specular reflectance may be available at early levels of vision, a range of compelling shiny and matte stimuli fail to produce efficient search. Therefore, present information suggests that shininess does not act as a basic feature. 2. Perhaps the necessary range of perceptual distance between stimuli wasn’t achieved with our stimuli (in spite of the fact that we used a variety of stimulus types). We remain on the lookout for better distractor sets that continue to control for confounding factors (e.g., motion, spatial frequency, color). 3. While generalized “shininess” does not seem to guide attention, it may be that more specific stimulus configurations like binocular luster can support efficient search. References Comments This project was funded by a grant from NEI Background: Wolfe & Franzel (1988) reported efficient search performance for stimuli defined by binocular luster... Monocular cues produce robust shininess percepts, even when contextual information is removed or replaced (Fleming, Dror, & Adelson, 2003; Hartung & Kersten, 2002). Back to the monocular world... from Fleming, R. W., Dror, R. O. and E. H. Adelson (2003) Does shininess guide attention? Moving Stimuli Our thanks to Roland Fleming, Ron Dror, Ted Adelson, Bruce Hartung, and Dan Kersten for allowing us to use their stimuli. Fleming, R. W., Dror, R. O. and E. H. Adelson (2003). Real world illumination and the perception of surface reflectance properties. Journal of Vision 3(5), 347-368. Hartung, B., and Kersten, D. (2002). Distinguishing shiny from matte [Abstract]. Journal of Vision, 2(7), 551a, http://www.journalofvision.org/2/7/551/, DOI 10.1167/2.7.551. Rensink, R.A. and Cavanagh P. (1994). Identification of Highlights in Early Vision. Investigative Ophthalmology & Visual Science,35:1623. [ARVO 1994; Sarasota, FL.] Wolfe, J. M. and Franzel, S. L. (1988). Binocularity and Visual Search. Perception & Psychophysics 44, 81-93. mean RT (ms) TARGET ABSENT TRIALS Errors (%) Set size n=14 mean RT (ms) Set size 0 1 2 3 4 5 6 7 8 9 shiny matte Stimuli: Spheres (4.5 dva) with either a white highlight (shiny) or dark patch (matte) were used. Disks (1 dva) containing the opposite dot polarity were placed just above the spheres to thwart search strategies dependent on locating a specific dot polarity in the display. A control condition was run in which observers searched for a specific configuration of light and dark patches on a flat surface. Experiment 1 TARGET PRESENT TRIALS Results: All searches were inefficient. It does not appear Os were searching for a particular configuration of light and dark patches. TARGET PRESENT TRIALS TARGET ABSENT TRIALS Shiny Set Matte Set Os searched for the presence of a member of their target set. Subjective ratings of shininess by a separate group of participants (n=11). 0 1 2 3 4 5 6 7 8 9 shiny matte Left Eye Right Eye by fusing these images, observers would perceive a salient lustrous patch where indicated in red. 250 350 450 550 650 750 850 950 1 2 3 4 0 10 20 30 40 50 60 70 80 90 100 shiny vs. matte y= 26x + 502 matte vs. shiny y = 44x + 515 mean RT (ms) Set size mean RT (ms) Set size 0 1 2 3 4 5 6 7 8 9 Shiny Sphere Rectangle Plain Sphere Subjective ratings of shininess by a separate group of participants (n = 11). TARGET PRESENT TRIALS TARGET ABSENT TRIALS 250 350 450 550 650 750 850 950 1 2 3 4 0 10 20 30 40 50 60 70 80 90 100 shiny vs. matte y = 22x + 451 matte vs. shiny y = 38x + 427 dot configuration y = 79x + 432 250 350 450 550 650 750 850 950 1 2 3 4 0 10 20 30 40 50 60 70 80 90 100 dot configuration y = 110x + 445 matte vs. shiny y = 38x + 491 shiny vs. matte y = 45x + 434 250 350 450 550 650 750 850 950 1 2 3 4 0 10 20 30 40 50 60 70 80 90 100 y= 67x + 484 shiny vs. matte matte vs. shiny y = 93x + 519 n=16 200 300 400 500 600 700 800 900 4 6 8 0 10 20 shiny sphere 200 300 400 500 600 700 800 900 4 6 8 0 10 20 y = 15x + 476 plain sphere y = 6x + 389 rectangle y = -x + 455 shiny sphere y = 27x + 492 plain sphere y = -2x + 501 rectangle y = 4x + 476 Set size mean RT (ms) Errors (%) Errors (%) Errors (%) Errors (%) Errors (%) n=15 mean RT (ms) Set size 300 500 700 900 1100 1300 1500 3 6 9 0 10 20 dot configuration 1 y = 45x + 628 dot configuration 2 y = 42x + 647 matte vs. shiny y = 10x + 573 shiny vs. matte y = 18x + 553 TARGET PRESENT TRIALS Errors (%) mean RT (ms) 400 600 800 1000 1200 1400 1600 0 10 20 Set size TARGET ABSENT TRIALS mean RT (ms) Errors (%) 3 6 9 Set size n=10 dot configuration 1 y = 94x + 606 dot configuration 2 y = 107x + 567 shiny vs. matte y = 41x + 609 matte vs. shiny y = 28x + 612 400 600 800 1000 1200 1400 1600 3 6 9 0 10 20 mean RT (ms) Set size TARGET PRESENT TRIALS Errors (%) n=15 700 900 1100 1300 1500 1700 1900 9 12 10 0 20 Errors (%) shiny vs. matte y = 41x + 808 matte vs. shiny y = 26x + 790 Results: Both searches were inefficient. shiny vs. matte y = 107x + 867 matte vs. shiny y = 94x + 881 TARGET ABSENT TRIALS mean RT (ms) 3 6 9 Set size 3 2 4 1 3 2 4 1 green = the critical condition Vision Sciences Society 2004 a sample search display (2/3rds original size) a sample search display (1/2 original size) shiny sphere rectangle plain sphere search for: among: a sample search display a sample search display (1/2 original size) search for: among: control search for: among: a sample search display a sample search display (1/2 original size) search for: among: a sample search display (1/2 original size) search for: among: control

Transcript of Is Shininess A Basic Feature In Visual Search? F111 · Is Shininess A Basic Feature In Visual...

Page 1: Is Shininess A Basic Feature In Visual Search? F111 · Is Shininess A Basic Feature In Visual Search? Randall S. Birnkrant Jeremy M. Wolfe Melina A. Kunar Matthew Sng A limited set

Is Shininess A Basic Feature In Visual Search?Randall S. Birnkrant Jeremy M. Wolfe Melina A. Kunar Matthew Sng

Brigham & Women's Hospital Brigham & Women's Hospital & Harvard Medical School Research Science Institute studentA limited set of attributes (e.g. color, size, and a varietyof depth cues) can guide the deployment of attention invisual search. Do monocular cues to shininess serveas guiding attributes?

INTRODUCTION

Rensink & Cavanagh (1994) found rapid search for acylinder with a top highlight, suggesting information aboutspecular reflectance is available at early levels of vision.

CONCLUSIONS

General MethodObservers (Os) were told to respond quickly and accurately. They were alsoinstructed to minimize eye movements. Stimuli remained on-screen until aresponse was made. Targets were present in approximately 50% of trials.Searches were blocked by target.

Does shininess guide attention? Static Stimuli

Experiment 2

Experiment 3

Experiment 5

Subjective ratings ofshininess by aseparate group ofparticipants (n=11).

Experiment 4

Stimuli:Shiny spheres rendered using photographicallycaptured real-world illuminations were takenfrom the work of Fleming, Dror, & Adelson(2003). Matte spheres were created by carvingup the specular reflections and replotting them.Clearly, the latter distorts global shapeinformation. However, this is only a potentialconfound if these searches proved to beefficient.

Results:Both searches provedinefficient. Shiny target #1did produce a noticeablyshallow search slope (7ms/item, shininess rating:8) and deserves furtherinvestigation.

Stimuli:Os searched for targets defined byconjunctions of shape (spherical orrectangular) and shininess. Thus, a searchtarget could either be a plain sphere, a shinysphere, or a rectangle cropped from thesame spherical specular reflection. Ifshininess is a feature, shininess-shapeconjunctions should yield efficient searches.

F111

Results: SEARCH FOR THE CONJUNCTIONOF SHININESS AND SHAPE WAS NOTEFFICIENT. Its not that these stimuli are toocomplex or noisy to support efficient search.Shiny rectangles among shiny and mattespheres popped out, as did matte spheresamong two types of shiny objects. However,observers could not use the combination ofshininess and shape to guide attention.

Stimuli:Similar to Experiment 1, shiny and matte sphereswere distinguished by either a white highlight ordark patch. The spheres were now textured(rotating uniform spheres are pretty dull) and oscil-lated through an angle of ~ 70 deg at 3 Hz. Again,disks containing the opposite dot polarity wereplaced just above the spheres to thwart searchstrategies dependent on locating a particular dotpolarity in the display. Control conditions were runin which Os searched for a specific configuration oflight and dark patches on a flat surface.

Results:Although target present RT x Set size slopes werequite shallow, SEARCHES WERE NOT FEATURE-LIKE. Again, it does not appear that Os weresearching for a particular configuration of light anddark patches.

Stimuli:Hartung & Kersten (2002) created compelling shiny andmatte “kettles” where surface appearance was defined bythe motion of specular reflections. Shiny kettles haddynamic specular reflections, whereas matte kettles hada “sticky” reflection - a single instance of specular reflec-tion that rotated with the kettle. Stimuli oscillated throughan angle of ~ 30 deg at 0.6 Hz. Rotational phase was ran-domly chosen for each stimulus to avoid grouping effects.The kettles did oscillate about different axes, but this isonly a potential confound if the searches proved to be effi-cient.

1. Although monocular information regarding specularreflectance may be available at early levels of vision, arange of compelling shiny and matte stimuli fail toproduce efficient search. Therefore, presentinformation suggests that shininess does not act as abasic feature.

2. Perhaps the necessary range of perceptual distancebetween stimuli wasn’t achieved with our stimuli (inspite of the fact that we used a variety of stimulustypes). We remain on the lookout for better distractorsets that continue to control for confounding factors(e.g., motion, spatial frequency, color).

3. While generalized “shininess” does not seem toguide attention, it may be that more specific stimulusconfigurations like binocular luster can support efficientsearch.

References

Comments

This project was funded by a grant from NEI

Background: Wolfe & Franzel (1988) reported efficientsearch performance for stimuli defined by binocular luster...

Monocular cues produce robust shininess percepts, evenwhen contextual information is removed or replaced(Fleming, Dror, & Adelson, 2003; Hartung & Kersten, 2002).

Back to the monocular world...

from Fleming, R. W., Dror, R. O. and E. H. Adelson (2003)

Does shininess guide attention? Moving Stimuli

Our thanks to Roland Fleming, Ron Dror, Ted Adelson, Bruce Hartung, andDan Kersten for allowing us to use their stimuli.

Fleming, R. W., Dror, R. O. and E. H. Adelson (2003). Real world illumination and the perception of surface reflectance properties.Journal of Vision 3(5), 347-368.

Hartung, B., and Kersten, D. (2002). Distinguishing shiny from matte [Abstract]. Journal of Vision, 2(7), 551a,http://www.journalofvision.org/2/7/551/, DOI 10.1167/2.7.551.

Rensink, R.A. and Cavanagh P. (1994). Identification of Highlights in Early Vision. Investigative Ophthalmology & VisualScience,35:1623. [ARVO 1994; Sarasota, FL.]

Wolfe, J. M. and Franzel, S. L. (1988). Binocularity and Visual Search. Perception & Psychophysics 44, 81-93.

mea

nR

T(m

s)

TARGET ABSENT TRIALS

Errors

(%)

Set size

n=14

mea

nR

T(m

s)

Set size

0123456789

shiny matte

Stimuli:Spheres (4.5 dva) with either a white highlight(shiny) or dark patch (matte) were used. Disks(1 dva) containing the opposite dot polarity wereplaced just above the spheres to thwart searchstrategies dependent on locating a specific dotpolarity in the display. A control condition wasrun in which observers searched for a specificconfiguration of light and dark patches on a flatsurface.

Experiment 1TARGET PRESENT TRIALS

Results:All searches were inefficient. It does notappear Os were searching for a particularconfiguration of light and dark patches.

TARGET PRESENT TRIALS TARGET ABSENT TRIALS

Shiny Set

Matte Set

Os searched for the presence of a member of their target set.

Subjective ratingsof shininess by aseparate group ofparticipants (n=11).

0123456789

shiny matte

Left Eye Right Eye

by fusing these images, observers would perceive a salient lustrous patch where indicated in red.

250

350

450

550

650

750

850

950

1 2 3 4

0

10

20

30

40

50

60

70

80

90

100

shiny vs. matte

y= 26x + 502

matte vs. shinyy = 44x + 515

mea

nR

T(m

s)

Set size

mea

nR

T(m

s)

Set size

0

1

2

3

4

5

6

7

8

9

Shiny Sphere Rectangle Plain Sphere

Subjective ratings ofshininess by aseparate group ofparticipants (n = 11).

TARGET PRESENT TRIALS TARGET ABSENT TRIALS

250

350

450

550

650

750

850

950

1 2 3 4

0

10

20

30

40

50

60

70

80

90

100

shiny vs. mattey = 22x + 451

matte vs. shinyy = 38x + 427

dot configurationy = 79x + 432

250

350

450

550

650

750

850

950

1 2 3 4

0

10

20

30

40

50

60

70

80

90

100dot configurationy = 110x + 445

matte vs. shinyy = 38x + 491

shiny vs. mattey = 45x + 434

250

350

450

550

650

750

850

950

1 2 3 4

0

10

20

30

40

50

60

70

80

90

100

y= 67x + 484shiny vs. matte

matte vs. shinyy = 93x + 519

n=16

200

300

400

500

600

700

800

900

4 6 80

10

20

shiny sphere

200

300

400

500

600

700

800

900

4 6 80

10

20

y = 15x + 476

plain spherey = 6x + 389

rect angley = -x + 455

shiny spherey = 27x + 492

plain spherey = -2x + 501

rect angley = 4x + 476

Set size

mea

nR

T(m

s)

Errors

(%)

Errors

(%)

Errors

(%)

Errors

(%)

Errors

(%)

n=15

mea

nR

T(m

s)

Set size

300

500

700

900

1100

1300

1500

3 6 90

10

20

dot configuration 1y = 45x + 628

dot configuration 2y = 42x + 647

matte vs. shinyy = 10x + 573

shiny vs. mattey = 18x + 553

TARGET PRESENT TRIALS

Errors

(%)

mea

nR

T(m

s)

400

600

800

1000

1200

1400

1600

0

10

20

Set size

TARGET ABSENT TRIALS

mea

nR

T(m

s)

Errors

(%)

3 6 9Set size

n=10

dot configuration 1y = 94x + 606

dot configuration 2y = 107x + 567

shiny vs. mattey = 41x + 609

matte vs. shinyy = 28x + 612

400

600

800

1000

1200

1400

1600

3 6 90

10

20

mea

nR

T(m

s)

Set size

TARGET PRESENT TRIALS

Errors

(%)

n=15

700

900

1100

1300

1500

1700

1900

9 12

10

0

20 Errors

(%)

shiny vs. mattey = 41x + 808

matte vs. shinyy = 26x + 790

Results:Both searches were inefficient.

shiny vs. mattey = 107x + 867

matte vs. shinyy = 94x + 881

TARGET ABSENT TRIALS

mea

nR

T(m

s)

3 6 9Set size

32 41

32 41

green = the crit ical condit ion

Vision Sciences Society2004

a sample search display(2/3rds original size)

a sample search display(1/2 original size)

shiny sphere

rectangle

plain sphere

search for: among:

a sample search display

a sample search display(1/2 original size)

search for: among:

control

search for: among:

a sample search display

a sample search display(1/2 original size)

search for: among:

a sample search display(1/2 original size)

search for: among:

control