Visual search based on motion information in natural images(Summary of Awarded Presentations at the 24th Annual Meeting)

(1)

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

The

J!panese

lournat

of

Ps,vchonomic

Science

2006,

VoL

25,

No.

1,

le3-le4

Summaryof

Awarded

PresentationIP07

Visual

search

based

on

motion

information

in

natural

images

Tamami

SuGIyAMA*･

**,

_Tatsuto

_TAKEucHI**,

_and

_Hisato

IMAI*

Toleyo

VVOman

's

Christian

Uhiversity*

_and

NTT

Communication

Science

Laboratories**

Wc

conducted

a visual search

experiment

in

which

participants

detected

a

target

(a

moving

natural

image)

which

moves

in

the

opposite

direction

to

the

other

distractor

images.

A

set size effect,

that

is,

decrease

of accuracy as

the

number

of stimuli

in

the

display

increased,

was

observed.

In

another experiment

to

identify

a

factor

that

restricted

the

_{participants'}

performance

in

the

task,

we

used

moving random-dot

patterns

having

various

amount

of relative motion.

We

found

that

a

target

with zero relative motion

did

pop-out,

but

the

accuracy

decreased

as

the

amount of relative motion

increased.

We

estimated

the

strength

of

the

relative motion of

the

moving

natural

images

based

on

the

motion-energy

model,

and

found

a negative

correlation

between

the

accuracy

and

the

strength

of

relative

motion

in

the

natural

images.

These

results

suggest

that

the

relative

motion

is

the

primary

faetor

for

a visual

search

of

moving natural

images

based

on

direction

inforrnation.

Key

words: visual

search,

natural

image,

motion-energy model

Introduction

A

visual

search

is

a

task

to

tind

a

target

stimulus embedded

in

a

number

of surrounding

distractor

stimuli.

It

has

been

shown

that

motion

information

is

crucial

in

searching

for

a

target

stimulus,

Because

simple

stimuli

such

as

moving

gratings

or

random-dots

have

been

used

in

a number of

visual

scarch studies regarding

motion

information,

how

we

search

for

moving objects

in

a

natural

image

is

a

question

which remams,

In

our experiments,

we

used

natural

images

and

random-dot

_patterns

as

visual

stimulL

The

partici-pants

were

asked

to

find

a

target

image

based

on

the

difference

in

motion

direction

between

a

target

and

distractor

images.

We

concluded

that

the

strength

of

the

relative

motion contained

in

a

natural

image

could

determine

the

sensitivity

to

the

visual search

task

of moving

natural

images.

*

_Division

_of

_Psychology,

_Graduate

_School

_of

Humanities,

Tokyo

Wornan's

Christian

sity,

2-6-1

Zempukuji,

Suginami-ku,

Tokyo

167-8585

**

_NTT

_{Comrnunication}

Science

Laboratories,

3-1

Morinosato-Wakamiya,

Atsugi-shi,

gawa

243-O198

Copyright

2006.

The

_Japanese

Psychonomic

Society

Method

In

the

visual

search

task,

the

participants

were

instructed

to

detect

a

target

Ca

natural

image)

embed-ded

in

other natural

images

which

were

distractors,

The

target

image

and

distractor

images

were

_played

in

opposite

directions,

For

example,

if

a

target

image

was

played

in

the

forward

direction,

the

distractor

images

were

_played

in

the

reverse

direction,

The

size

of each

irnage

was

3,9

×

3.9 deg,

The

display

screen

was

divided

into

9

region$

which

did

not overlap,

Each

image

which was

shown

was

positioned

in

one of

the

regions.

In

a

single

triaL

4

or

9 images

were

displayed

on

the

screen,

The

task

of

the

participants

was

to

specify

the

location

of

the

target

image.

Half

of

the

trial

contained a

target

image,

and

the

other

half

of

the

trials

did

not

Each

moving

image

was

displayed

for

1 sec,

and

the

_participants

selected

the

location

of

the

target

image

by

clicking a mouse

button

after

the

termina-tion

of

the

moving

images.

When

they

did

not

detect

a

targer

image

they

selected

a

location

where

images

were net

displayed.

In

the

first

experiment

we used

9

moving natural

images:

such as

a

horse

running

on

the

ranch, or a

moving

_goldfish.

In

the

second experiment we usecl

random-dot

patterns

moving with

various

amounts

of relative motion.

The

amount

of relative

motion

All

rights reserved,

(2)

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

104 The

_Japanese

_Journal

of

Psychonomic

Science

Vol,

25,

No.

1 (a)90Aeh

8eoffsE

7o

6e

(b)

1OOn5>

geeege

so

7eINI=2

l

4secsize9

tht`ve

sc/iLta

iwhkft

Figure

1. Accuracy

of

visual

search

for

moving

natural

images

as a

function

of set size,

(b)

Accuracy

of

visual

search

for

moving random-dot

patterns

as

a

function

of

the

strength

of

relative

motion.

The

error

bars

represent

±

1 SE,

was

determined

by

the

_area

_ratio

_of

_a

_central

_part

and

a

_peripheral

_part

that

moved

in

opposite

direc-tions,

The

ratios were

O

:

16 (zero

relative

motionL

1

:

15 (small

relative

motion), and

4

:

11 (large

relative

motion),

Results

The

accuracy

(percent

correct} of

the

visual search

for

moving natural

images

as

a

function

of

the

set-size

is

shown

in

Figure

1(a)

which

illustrates

that

a significant

set-size

effect was observed.

With

mov-ing

random-dot

patterns

the

accuracy

of

target

de-tection

decreased

as

the

amount

of

relative

rnotion

increased,

as

shown

in

Figure

1(b},

Discussion

In

the

first

experiment

with

moving

natural

im-ages

we

observed

a

set-size

effect.

That

is,

the

visual

search

was essentially

difficult,

In

the

second

experi-ment

the

visual

search

task

with moving randem-dot

patterns

having

relative

motion was

more

diMcult

than

the

other

task,

The

results

suggest

that

in

the

first

experiment

a

factor

that

restricts sensitivity

could

be

_produced

from

relative motion

components

contained

in

moving natural

images,

To

examine

this

_possibility

we

estimated

_the

strength

of

the

relative

motion

of moving natural

images

based

on

the

motion-energy model

(Adelson

&

Bergen,

1985),

We

found

a

negative

correlation

between

the

accuracy

and

the

strength of relative

motien

in

the

natural

images

in

the

first

experiment,

We

conclude

that

the

relative rnotion

is

the

_primary

factor

for

a visual search

of

moving

natural

images.

Reference

Adelson,

E.

H. &

Bergen,

J. R,

1985

Spatiotemporal

energy

models

for

the

_perception

of

motion.

Ibur-nal