1P045 ハトにおける人の顔画像の認識 : 動画から静止画への転移(2003年度 日本基礎心理学会第22回大会優秀発表賞)

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

The _{JdPaneseJburnatof}PsJ,chonomic SciencE

2004,N'oL23,Ne.1,95･

-96

Summary

ofAwarded

PresentationIP045

Recognition

of

_pictures

of

human

faces

by

_pigeons

Transfer

from

dynamic

to

static

stimuli

i)

Hiroshi

MAKiNo*

and

Masako

JITsuMoRI*'

**

Chiba

Ubeiversily*

In

a go/no-go

discr[mination

procedure,

four

_pigeons were trained todiscriminate dynamic

irnages

of

four

human faceswhieh werc rotating around the _y-axisfrorn+670 to

-67".

After

completion of thetraining, transfertostatic images at nine

depth

orientations

_(OO,

±

23e,

±

45e,

±

670,

and ±

･

900)

was tested

in

extinction.

Thc

pigeons showed excellent transferto thestatic stimuli,

but

responses tothc _positivefacessignificantly decreased

for

novel views eutside the range spanned by the dynamic stimuli, The findings suggc$t that theresponses of thepigeons

were

based

on two-dimensional propertiesof thefacesseen inthe trainingimages rather than the

three-dimensional properties of the

human

faces,

Key words: dynamic stimuli,

face

recognition, _pigeon, rotation

invariance

Introduction

The retinaL image of an object varies as a

function

of orientations,

distances,

lighting

conditions,

back-ground scenes, and other aspects at thetime of

vicw-ing,Invarianee operations are required

for

observers

to

identify

objects

despite

a varinbility of retina]

stimulation. One of the rnost sophisticated

invari-ance operations

is

the

recognition of objects

from

novel viewpoints.

Jitsumori

& Makino

(in

presis,Experiment 1)

train-ed pigeons to discriminate

betwee.n

frontal

views of

human

faces

on a

CRT

monitor

in

a _golno-go

discrimination

_procedure. The _pigeons showed

sub-stantial _{generalization} to novel views, even though images of human faceschange radically as the view-pointchanges.

However,

the

discrimination

of novel views was not _perfect,and responses to the_positive faces

declined

systernatically wirh the

degree

of

rota-tion

from

the trainingview,

One

possible explanation was

that

the

_pigeons

ii_{A detailedversion isincluded in}

_Jitsumorl,

_&

Makino

(in

press)as a part ef

the

series of

experirnents reported in_the article.

*

_Graduate

_School

of

Sciencc

and Technology,

Chiba

University,

Yayoi-cho,

Inage-ku, Chiba

263-8522

** _{Department of Cognitivc & Information}

ences, Faculty of Letters,

Chiba

University,

Yayoi-cho,

Inage-ku,

Chiba

263L8522

perceived "three-dimensionai"

video

images

of

hu-man

faces

but generalization tothe facesat novel

viewpoints was dccr¢ased due to discrimination

be-tween

the

famiLiarand novel views.

In

the

_present

experiment, we trained

_pigeons

te

discriminate

im-ages of human faceswhich were dynamically rotat-ing in depth. Ifseeing a human facecontinuously

changing itsorientation

in

depth

enabled the

pi-geons to

integrate

the

different

views as a unified 3D

object, then transfer would occur not only to the

corresponding static views

but

also to the novel

views outside the range spanned by the dynamic

stimuli,

Method

Subjects

Four

experimentally naive

homing

pi-geons were used as subjects,

Apparatus

Two

conventienal operant chambers,

each of which

had

a screen

key

and a start

key,

were used. The stimuli were disp]ayed via a laservideo

disk

playertoa

CRT

color monitor.

Stimuli

Images of thefacesof four

undergradu-ate male students were used as stimulL A dynamic

sequence consisted of a face which rotated from

+67e _to

-670

_in

₅

seconds, so that the

face

in

the

displaycontinually rotated totheleftand right. Dur-ing _{genera]izatien}testing, static images of the faces

at nine depth orientations

(Oe,

±230, ±4sO, ±670,

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

96

.Itt'

'/･l,1A

I],/,iiik･ft.l/

_t

_ttA-23e

,me il'

,

,tsim･･A+23e

Figure

B

t

_t

11..t...tE/ 'ge;,\fll

-45e

The

_Japanese

_Journal

of

Psych

uat

'

i'1'

A+Ui"

1,

Example$

of

ductions of the stimuli.

1.25v-t2

i,oo8coe

o.7s$FS O.50Nz<.

O.25E

o.oo

-90 -67 -45 -23

e +23

DEGREES

OF

Figure 2, The rnean relative

posi,tivefaces

{open

circles) and

faces

{fi11ed

circles}

during

The

region between the

represents the range of the

used

for

training.

are shown inFjgure 1.

Procedure The su

tothe

dynamic

images of two faces

spond to thosc of the remaining

l.i'･

}･

c

'IY,･

A-67e

t

ttt/../

_{./..di1,,,;･i･}

s.

'

i,illl/l,,

A+67"

black-and-white

･/11-

･

tt

t/t

ti.aj'･Ilii･'i

//･-//･tt.

ttl

･

/a 1/

'

D

'

't'

_/g.

.1･1･1･/･1/･

.･t//,/

A-go"

es.1di'

,/I,/1

A+goe

+45 +67 +90

ROTATION

responses to the the negative generalization. vertical tines

dynamic

view

bjectswere trained torespond

but not to

two

faces

in

a

onomic

Science

VoL

23,

No.

1

go/no-go discrimination procedure similar to that

used by Vaughan and Greene

(1984},

A training session consisted of

30

positiveand

30

negative

tri-als, with an intertrialintervalof 5seconds, Training

continued until 80% or more of the totalrcsponses

occurred on positivetrials.

The

birds

then received

generalizationtestingwith thestatic stimulj, A set

of

36

dtfferent

static stimuli was _presented twice

in

a

session.

Two

sessions were _given inextinction.

Results

and

Discussion

Three

of the four _pigeons learnedthe

tion.

Figure

2

shows the mean relative

tion _gradients averaged across the three

_pige6ns,

The

pigeo]s showed an excellent transfer

from

the dynamic tothe static views over the ±67' range. But the responses tostatic

images

significantly

creased when the novel views

(

±

.900)

of thepositive

faces

were outside the range spanned by the

narnic stimulL

In

the

investigation

ei

Jitsumorj

and

Makino

(in

press),the pigeons which were trained

and testedwith static stimuli failedtoshow a

fer

tothe

dynamic

images.

This

finding,

together

with theobservation inthe_present experiment of a

failure

to

transfer

to

the

novel views, suggest that

pigeons are

insensitive

tothethrce-dimensional cues

in video images, Jt,istherefore

likely

that the geons- generalization responses were based on the

2-dimensional training images rather than the

3-dimensional properties of

human

faces,

References

Jitsumeri,

M.

&

Makino,

H,

in

press.

Recognition of

static and

dynamic

images

of

depth-rotated

human facesby pigeons.

Learning

&

Behavior.

Vaughan, W. & Greene, S.L. 1984 Pigeon visual

memory capacity.

fournat

of

ExPerimental

ogly:

Animal

Behavior

Processes,

10,256-271.