物体の内的表現における奥行き情報の寄与についての検討(第23回大会 優秀発表賞抄録)

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

lhefopaneseJbut7!alofFig),ehonomicScienca

2005,VoL 24,No. I,I]5-116

Summary

ofAwarded

PresentationIP16

The

role

of

depth

information

in

object

recognition

Ayako

SANEyosHi*,

Chikashi

MicHiMATA*,

and

Keiji

SANEyosHI**

Sophia Lbeiversity*and TohJ,o

instttute

of

_{Teehnol(lgy**}

"Geons"

(Biederman,

1987)

with differentrotatlon angles were sequentially _presented through

a

haploscope

and the_{participantsperformed} a sequential matching task.The RT increasedas the

rotation angle increased, The

3D

condition,

in

which thestimulus was stereoscopically _presented,

was _performed fasterthan the2D eondition.

The

eft'ect of the rotation angle was equivalent

for

the

2D and 3D conditions.

Furthermore,

thercwas a _possibilitythatdepth information

is

_{helpful in} extracting thc qualitative

information

of an object.

These

results suggested that the _qualitative

information

extracted frornview-dependent

2D

representation plays a crucial role inconstructing

therepresentation of objects.

Key words: object representation, view-dependency, stereoscoplc vision

Tarr,Williams, Hayward, and Gauthier

(1998)

re-ported thattherccognition of _geons was affected

by

achange ef viewpoint. Thus they concluded thatthc representation of un objcct would not be a

3D

view-independent

$tructural description

but

2D

view-dependent images. However, thc structural

descrip-tionmodel suggested thatrhe

depth

information of

an object would be very important for

reconstruct-ingthe representation of a

3D

objcct.

Because

Tarr

et aL

(1998)

employed 2D

images

as stimuli, thereis

a possibilitythat theirresult was due to a lack of

depth

iniormation

of the objects, In _{the present}

experiment we investigatedtheeffect ofdepth

infor-mation in recognizing _geons

from

_different

view-points.

We

presented thegeons stereoscopicaily

us-inga haploscope.

Methods

Participants

Sixteen graduate and

undergradu-ate students served as volunteer participants.

Apparatus

The

stimuli were _presented on a

SONY

19-inch

CRT

rnonitor connected

to

an

Apple

PowerBook G4 computer, A ten-key

_pad

was

con-nected to the computer and served as a two-key

response console.

Through

a

haploscope

each eye

views itsrespective halfside of themonitor

Ce.g.

_the

*

_Department

_{of Psychology,}

_Sophia

_University,

1 Kioicho,

Chiyoda-ku.

Tokyo,

102-8554

** _{Radioisotope Research}

_Center,

_{Tokyo Institute}

of Technology,

Nagatsutacho,

_Midori-ku,

harna-shi,

Kanagawa

226-8501

left

eye sees the

left

ha]fof monitor).

In

a

2D

condi-tionboth eyes receive identical

images.

In

_the

pres-cnt investigationwhen the 3D condition was used

two

images

with

binocular

disparitywere _prcsented.

Stimulus The stimuli were

20

_gray-shaded

_geons

{Biederman,

1987)

which were rotated around the

vcrtical axis with a step of 15 degrees. All of the

stimu!i were appreximately 6X6 degrees of visua] angle insize.

Design

Theexperirnentaldesignwasanorthogo-nal combination of "View"

(2D

vs.

3D)

and "Rotation

Angle"

(Oe,

150,30a,45",600,

750,

and

900).

Both

of

the variables were manipulated within the

partici-pants.

Task andprocedure Theparticipants were asked

to decide whether

2

successively presented stimuli

were the same _geons, ignoring theirrotation. In2

scparate sessions the _partjcipants observed the

geons with a haploscope in

both

_the

2D

and 3D

condition. Ineach triala

fixation

cross appeared for

500

ms, and thiswas

followed

by a 200 ms

presenta-tionof the

first

stimulus. A mask stimulus was then

displayedfor500 ms, and thiswas

followed

by

a 1OO

ms presentation of the second stimulus.

Results

and

Discussion

The

results are shown inTable

1(a).

The

reaction

time data was subjected toa

2

(View)X7

_(Rotation

Angle) analysis of variance with repeated

measure-ments.

The

main effect of Rotation Angle was

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

116

The

Japanese

Journal

oi Psychonomic

Science

Vol.

24,

No.

1

nificant

(F(6,

90)=522,

pex',Ol).

As

can be seen

in

Table 1a,

RT

linearly

increased as the Rotation

An-gle

increased,

both inthe2D and 3D conditions. The

main effect of View was also significant

(F(1,

15)=

4.92,

_p<.05),

indicatingthat the

3D

condition was

performed with shorter

RTs

than the 2D condition.

The interactionof theRotation Angle by View was

not significant. ThaL is,

the

Rotation

Angle

was

equally affected inthe 2D and 3D conditions. This

implies

that the view-dependency in recognition of

geons, as reported by Tarr et aL

(1998),

cou]d nol

be

cxplained b},a lack of depth infoTmation for the

objects.

The

present results support the theory that

theobject representation isbased on vicw-depcndent

2D images.

On

the

other

hand,

it

should

be

noted thatthe

3D

condition was pcrforrned with a shorter RT than the

2D

condition. Itisthereforenecessary toconsidcr

t.hcroLe of depth information in constructing

2D-based representation.

Hayward

and Tarr

(1997)

sug-gestedthattheeffect of rotatien was largerwhcn the

rotation clid not _{preserve qua]itative} information

(e.g.

the main axis wu' scurved or stra{ght) than when

it

did

not preserve quantitative information

(e.g.

the

curvature of the main axis s-,as changed, see Figure

1)but did _{preserve qualitative}information. Thev

concLuded _{that qualitative}

information

would be

more

important

than _quantiLative jnformation for

construction of a representation of theobject.

XVethercfore considered the _{possibilitythat}

qua]i-tativeinformation could

be

extracted much easier

when depth information was added and we

re-ana]yzcd the data

to

examine

this

_possibility.

The

RT data was subjected to a

2

_(View:

2D, 3D)X2

{Cha]ge

Type: _{quantitative,qualitative;}see

Figure

1)

analysis of variance with repeated measurements,

The results are shown in Table 1(b). The main

effect ef Change Type was not statistically

sig-nificanL

(F(1,

15)==3,35,

P=.09). The interactionof

View

by Change Type was also not statistically

sig-nificant

(F(1,

15)=3.23,

P=.09),

but

_{the quantitative}

change _produced shorter RTs than the qualitative

information

changc

in

the 2D condition, cSidakistesti

P<.05). However a significant differenccwas not

Table 1. The mean RTs,

(a)

View by Rotation

Angular

Type

Oo150

300

45e

600

750

900 Mean

2D 457 470 475 481 480 477 486 475

3D 439 456

453

460

455

475

474

458

(b)

View by Change Typc

Quantitativechange

(ms)

Qualitative

change

2D3D

465466

481464(ms)

Figure

1.

An

example of the changed typc

(Change

Type) of stimuli,

revealed inthe _quantitative change and _qualitativc

change in the3D condition,

(Sidak's

test:

p=.69).

Consequently,

the addition of depth information

reducecl the cost of the _qualitat.ivechange.

This

result

implies

that

depth

{nformatjon

is

helpful

_for

extracting the quaritativc[nformation of an object.

These

results

indicated

that

the

_qua]itative

informa-tion extracted from view-dcpendent 2D images _plays a crucial role inconstructing theobject

representa-tion,

References

Biederman, I.1987 Recognition-by-components: a

theory

of

human

image

understanding.

Rsychotogi-catReview,

94,

115-

147.

Hayward, W,

G,,

& Tarr,M.

J,

1997 Testing

tions

for

viewpoint Lnvariance jnobject

tion.

fournal

of

iixPerimentaZ

Ps.vchotog:y:

HPR

23,.

1511-152L

Tarr,

M.

J.

Williams,

P.

Hayward,

W,

G,,

& Gauthier,

L

1998

Three-dimensional

object recognition is