Keywords: visual recognition, time course, detection, categorization, identification

Visual Object Detection, Categorization, and Identification Tasks Are
Associated With Different Time Courses and Sensitivities

Stephan de la Rosa, Rabia N. Choudhery, and Astros Chatziastros
Max-Planck-Institute for Biological Cybernetics, Tübingen, Germany

Recent evidence suggests that the recognition of an object’s presence and its explicit recognition are
temporally closely related. Here we re-examined the time course (using a fine and a coarse temporal
resolution) and the sensitivity of three possible component processes of visual object recognition. In
particular, participants saw briefly presented (Experiment I to III) or noise masked (Experiment IV) static
images of objects and non-object textures. Participants reported the presence of an object, its basic level
category, and its subordinate category while we measured recognition performance by means of accuracy
and reaction times. All three recognition tasks were clearly separable in terms of their time course and
sensitivity. Finally, the use of a coarser temporal sampling of presentation times decreased performance
differences between the detection and basic level categorization task suggesting that a fine temporal
sampling for the dissociation of recognition performances is important. Overall the three probed
recognition processes were associated with different time courses and sensitivities.

Keywords: visual recognition, time course, detection, categorization, identification

According to many influential theories, visual object recognition
is not a unitary process but consists of several component pro-
cesses that are carried out in some temporal (e.g. Marr &
Nishihara, 1978; Biederman, 1987; Nakayama, He, & Shimojo,
1995). Detection, basic-level categorization, and identification are
considered to be candidate component processes of object recog-
nition (see e.g. Grill-Spector and Kanwisher, 2005). Here we
define detection as the observer’s judgments about an object’s
presence. Furthermore we refer to categorization as the recognition
of the object’s basic-level category (e.g. dog) and to identification
as the recognition of the object’s subordinate category (e.g. Ger-
man Shepherd; Rosch, Mervis, Gray, Johnson, & Boyes-Braem,
1976).

A recent debate concerns the temporal in which these
component processes of object recognition are executed. A com-
mon view is that objects or object features are detected in the
background before they are recognized in more detail (e.g.
Nakayama et al, 1995). Recent evidence, however, suggests that
the detection of an object and its categorization are associated with
very similar reaction times. These results lead to the suggestion
that the visual processes underlying detection and categorization
are equally fast (Grill-Spector & Kanwisher, 2005).

According to the first view and several “classic” theories of
object recognition, the visual system first detects parts of an object

(e.g. object contours). Subsequently these parts are then integrated
into an object representation which serves as a basis for a more
detailed analysis of the object (e.g. Marr, 1982; Biederman, 1987;
Nakayama et al., 1995; see also Treisman & Gelade, 1980). This
kind of organization in which the detection precedes categorization
or identification allows object recognition to be more efficient
because processes underlying object categorization or identifica-
tion can focus on visual information that pertains to an object
rather than processing the entire visual array. According to this
view visual processes underlying detection temporally precede
visual processes underlying categorization. Hence one would ex-
pect that detection is associated with shorter reaction times than
categorization or identification.

More recent evidence challenges the view that detection
temporally precedes categorization. Grill-Spector and Kan-
wisher (2005) measured participants’ reaction times and accu-
racy to detect, categorize, and identify images of objects in their
natural background for various presentation times. Interest-
ingly, they found that both accuracy and reaction times did not
differ significantly between object detection and object catego-
rization at all tested presentation times. Yet, object identifica-
tion was clearly associated with significantly lower accuracy
and significantly higher reaction times for all presentation
times. Additionally when Grill-Spector and Kanwisher (2005)
asked participants to both detect and categorize an object on a
single trial, a trial-by-trial analysis revealed that categorization
errors were related to detection errors and vice versa. Grill-
Spector and Kanwisher (2005) therefore concluded that object
detection and object categorization have the same time course.
In contrast identification has a time course that is shifted
towards longer reaction times suggesting that its underlying
visual processes are slower.

Using the same experimental design, Mack, Gauthier, Sadr, &
Palmeri (2008) only partly replicated Grill-Spector and Kanwish-

This article was published Online First November 1, 2010.
Stephan de la Rosa, Rabia N. Choudhery, and Astros Chatziastros,

Department of Human Perception, Cognition, and Action, Max-Planck-
Institute for Biological Cybernetics, Tübingen, Germany.

This research was supported in by the EU Project “Joint Action Science
and Technology” (IST-FP6-003747).

Correspondence concerning this article should be addressed to Stephan
de la Rosa, Department of Human Perception, Cognition and Action,
Max-Planck-Institute for Biological Cybernetics, Spemannstr. 38, 72076
Tübingen, Germany. E-mail: [email protected]

Journal of Experimental Psychology: © 2010 American Psychological Association
Human Perception and Performance
2011, Vol. 37, No. 1, 38 – 47

0096-1523/10/$12.00 DOI: 10.1037/a0020553

38

T
hi

s
do

cu
m

en
t i

s
co

py
ri

gh
te

d
by

th
e

A
m

er
ic

an
P

sy
ch

ol
og

ic
al

A
ss

oc
ia

tio
n

or
o

ne
o

f i
ts

a
lli

ed
p

ub
lis

he
rs

.
T

hi
s

ar
tic

le
is

in
te

nd
ed

s
ol

el
y

fo
r t

he
p

er
so

na
l u

se
o

f t
he

in
di

vi
du

al
u

se
r a

nd
is

n
ot

to
b

e
di

ss
em

in
at

ed
b

ro
ad

ly
.

er’s (2005) results. They showed that object detection and catego-
rization are tightly temporally coupled when images were pre-
sented upright. However, when the same objects images were
presented upside-down, participants’ detection and categorization
performance was significantly different. In particular, the detection
of inverted objects was significantly better than their categoriza-
tion as indicated by larger d� values for detection. Furthermore,
participants exhibited faster response times in the detection task
than in the categorization task when object images were inverted.
Mack et al. also found the detection of degraded (i.e. phase
scrambled) object images to be faster and more accurate than their
categorization. These results suggest that under more challenging
viewing conditions object detection and categorization perfor-
mance can be dissociated. However, the finding that detection and
categorization of upright object images are associated with the
same time course remains unchallenged.

The finding that detection and categorization are tightly tempo-
rally linked for upright object images imposes important con-
straints onto existing theories of object recognition. Here we
re-investigated the time course of detection and categorization of
upright natural object images (Grill-Spector & Kanwisher, 2005).
The close temporal linkage between detection and categorization
might be owed to the rapid nature of visual categorization (Thorpe,
Fize, & Marlot, 1996; VanRullen & Thorpe, 2001). Hence, if
differences between detection and categorization performance ex-
ist, they should occur at very short presentation times. We there-
fore decided to use a finer temporal sampling at short presentation
times than in previous studies to examine the time course of visual
recognition.

Experiment I: The Time Course of Object Recognition

We measured the time course of object detection, categoriza-
tion, and identification using a finer temporal resolution for short
presentation times than in previous studies (Grill-Spector & Kan-
wisher, 2005; Mack et al., 2008). Moreover, we were interested in
assessing the degree of detail that participants could perceive from
a single brief presentation of an image by probing detection,
categorizing, and identification on the same trial. If detection is
mediated by faster visual processes than categorization, then par-
ticipants should be able to tell the presence of an object while
being unable to categorize it at short presentation times. To this
end, participants saw two temporally separated, backward-masked
image presentations with one showing an object image (e.g. a dog)
and the other a non-object image (visual noise; see Figure 1).
Following these two image presentations participants had to indi-
cate on an answer screen the object’s presentation interval (detec-
tion task), the object’s basic-level category (categorization task),
and the object’s subordinate category (identification task).

Methods

Participants. Ten naı̈ve participants (age range between 18
and 30 years; four females) participated in the experiments. All
participants had a normal or corrected-to-normal vision and were
naı̈ve to the task and stimuli. Each participant gave informed
consent prior to the experiment and was compensated 8€/hour for
their participation.

Apparatus and stimuli. Stimuli were presented on a Sony
(Tokyo, Japan) Monitor (CPD-G500) by means of the Psychtool-

box (Brainard, 1997; Pelli, 1997). The gamma corrected monitor
had a refresh rate of 140 Hz.

The grayscale object images were drawn from six object cate-
gories (bird, boat, car, dog, house, and flower). Each category
contained 100 images. Within each of the six categories, 50 images
were of a particular subordinate category, hereafter referred to
exemplar. The six exemplars for the six categories were pigeon,
sailboat, VW Beetle, German Shepherd, barn, and rose. The other
50 images in a category were non-exemplar images, that is, images
of a different subordinate category than the exemplar category
(e.g. all other birds except pigeon). The non-object images were
patches of Gaussian visual noise. A new patch of visual noise was
used on each trial. The mask was a scrambled version of an object
image. To do so each object image was chopped up into 10 pixels
by 10 pixels tiles that were then randomly rearranged. All stimuli
were presented in the center of the screen with a gray level of 127
pixel (RGB value). All stimuli had the same size (5.89° visual
angle), luminance (127 RGB pixel value), and contrast (20 RMS
RGB pixel contrast).

Procedure. A trial began with the start screen “click here &
watch” presented in the centre of the screen (see Figure 1). Par-
ticipants had to left click with the mouse on “click here & watch”
to start the trial. Following the mouse click, the start screen was
replaced with a gray screen for 50 ms to minimize forward mask-
ing. The gray screen was followed by two image presentation
intervals which were separated by 500 ms inter-stimulus-interval
consisting of a gray screen. One interval presented a real-object
image (an object from one of the categories) and the other interval
presented a non-object image (visual noise). In both intervals the

Figure 1. Schematic outline of an experimental trial in Experiment I. The
labels to the right of a screen indicate the presentation time (in ms) of the
corresponding screen. The mouse symbols indicate the screens for which
participants mouse clicks were required to continue to the next screen, i.e.
to start a trial and to respond (selected answers were highlighted in white).
The answer screen is shown enlarged along with the correct answers for
this trial for sake of clarity.

39TIME COURSE AND SENSITIVITY OF VISUAL RECOGNITION

T
hi

s
do

cu
m

en
t i

s
co

py
ri

gh
te

d
by

th
e

A
m

er
ic

an
P

sy
ch

ol
og

ic
al

A
ss

oc
ia

tio
n

or
o

ne
o

f i
ts

a
lli

ed
p

ub
lis

he
rs

.
T

hi
s

ar
tic

le
is

in
te

nd
ed

s
ol

el
y

fo
r t

he
p

er
so

na
l u

se
o

f t
he

in
di

vi
du

al
u

se
r a

nd
is

n
ot

to
b

e
di

ss
em

in
at

ed
b

ro
ad

ly
.

presentation of an image was immediately followed by a mask
(scrambled object image) that was visible for 500 ms. Following
the image presentation the answer screen was presented. It always
presented the same three questions along with the same answer
options (see Figure 1). The three questions were designed to
measure detection, categorization, and identification, respectively.
Participants answered all three questions by selecting one answer
option for each question with a mouse click. Participants were
instructed that a) they may answer the questions in any ; b)
once an answer was selected, it could not be changed; c) all three
questions are of equal importance; and d) they should guess an
answer if they did not know the answer to a question. Participants
had to answer all three questions to move on to the next trail. Once
the three questions were answered, the next trial started by pre-
senting “click here & watch” in the centre of the screen.

Forty-two trials constituted a block, and seven blocks an exper-
iment (total of 294 trials). The real-object was pseudo randomly
assigned to one of the two intervals on each trial with the restric-
tion that the real-object had to appear in the first and second
presentation interval equally likely (50%) within a block. The
probability of guessing the correct answer of the detection question
was therefore p � .5. Seven images from each of the six categories
were shown in a block. Hence the probability of guessing the
correct answer of the categorization question was p � 1/6 (six
categories). Out of the seven images that were presented of a given
category, six were exemplar images and one was a non-exemplar
image. Hence, in total six non-exemplar images were shown
within a block across all six categories. The probability of cor-
rectly guessing the correct answer (six target exemplars names
plus the option “other” [see top right side of Figure 1]) of the
identification question was therefore p � 1/7. Each object image
was presented only once. Both, the object and the non-object
images, were presented for the same duration. The presentation
time was randomly selected on a given trial from the following

presentation times: 7, 21, 28, 36, 57, 78, or 121 ms. Each presen-
tation time was used six times (i.e. the frequency with which each
presentation time occurred was counterbalanced) within a block
and 42 times during an experiment. That is, the presentation
of presentation times was randomized while the presentation fre-
quency was counterbalanced across presentation times.

Results and Discussion

Recognition performance was measured in terms of corrected-
for-guessing accuracy scores using the following formula (Mac-
millan & Creelman, 2005, p. 252):

c � �m � p�c� � 1�/�m � 1� � 100, (1)

where c is the accuracy corrected for guessing in percent, p(c) is
the probability of a correct response, m is the number of answer
alternatives in a given task; m � 2 in the detection task, m � 6 in
the categorization task, and m � 7 in the identification task.

Figure 2 left panel shows the psychometric functions relating
accuracy and presentation time for each of the three recognition
tasks separately. The psychometric functions for detection, cate-
gorization, and identification have clearly different shapes. A
repeated-measures analysis of variance (ANOVA) with presenta-
tion times and recognition tasks as within-subject factors was
conducted to investigate whether the observed differences in Fig-
ure 2 left panel bear statistical significance. Both main effects of
presentation time, F(6, 54) � 146.17, p � .001, �partial

2 � 0.942,
and recognition task, F(2, 18) � 27.30, p � .001, �partial

2 � 0.752,
were significant. The interaction of presentation time and recog-
nition task was also significant suggesting that the presentation
time had a different effect on accuracy scores for the three recog-
nition tasks, F(12, 108) � 23.70, p � .001, �partial

2 � 0.725. To see
which recognition tasks differed from each other, we conducted
two separate (detection vs. categorization and categorization vs.

0 100 200
0

10

20

30

40

50

60

70

80

90

100

Presentation time (ms)

co
rr

e
ct

e
d

−
fo

r−
g

u
e

ss
in

g
a

cc
u

ra
cy

detection
categorization
identification

0 100 200
−20

−10

0

10

20

30

40

50

60

70

Presentation time (ms)

a
cc

u
ra

cy
d

iff
e

re
n

ce
(

%
)

0 100 200
−20

−10

0

10

20

30

40

50

60

70

Presentation time (ms)

a
cc

u
ra

cy
d

iff
e

re
n

ce
(

%
)

detection−categorization categorization−identification

Figure 2. Results of Experiment I. Left: Psychometric functions relating mean accuracy (corrected-for-
guessing) to presentation time (in ms) for each of the three recognition tasks in Experiment I. Bars indicate one
standard error of the mean. Middle: Mean accuracy difference between detection and categorization performance
calculated for each presentation time separately. Right: mean accuracy difference between categorization and
identification calculated for each presentation time separately. Bars indicate 95% Bonferroni-corrected confi-
dence intervals in the middle and the right panel.

40 DE LA ROSA, CHOUDHERY, AND CHATZIASTROS

T
hi

s
do

cu
m

en
t i

s
co

py
ri

gh
te

d
by

th
e

A
m

er
ic

an
P

sy
ch

ol
og

ic
al

A
ss

oc
ia

tio
n

or
o

ne
o

f i
ts

a
lli

ed
p

ub
lis

he
rs

.
T

hi
s

ar
tic

le
is

in
te

nd
ed

s
ol

el
y

fo
r t

he
p

er
so

na
l u

se
o

f t
he

in
di

vi
du

al
u

se
r a

nd
is

n
ot

to
b

e
di

ss
em

in
at

ed
b

ro
ad

ly
.

identification) within-subjects ANOVAs with presentation time
and recognition task as within-subject factors.

Detection vs. Categorization

Figure 2 (left panel) suggest that the psychometric function for
detection and categorization clearly differ. A within subject ANOVA
with presentation time and recognition task as within subject factors
was used to investigate the statistical significance of this observation.
Both main effects of presentation time, F(6, 54) � 120.68, p � .001,
�partial

2 � 0.931, and recognition task, F(1, 9) � 18.67, p � .002,
�partial

2 � 0.675, were significant. The significant interaction of
presentation time and recognition task indicates that detection was
better than categorization only on some presentation time levels,
F(6, 54) � 34.75, p � .001, �partial

2 � 0.794. Bonferroni-corrected
paired t-tests conducted for each presentation time separately re-
vealed that detection performance was better than categorization
performance for presentation times at 21, 28, and 36 ms (see
Figure 2 middle panel).

Categorization vs. Identification

Figure 2 (left panel) suggests performance differences between
the categorization and identification task. A two-way, within-
subjects ANOVA with presentation time and recognition task as
factors was used to examine these observed differences. We found
a significant main effect of presentation time, F(6, 54) � 155.92,
p � .001, �partial

2 � 0.945, and recognition task, F(1, 9) � 17.37,
p � .002, �partial

2 � 0.659. The interaction of presentation time and
recognition task was also significant, F(6, 54) � 3.35, p � .007,
�partial

2 � 0.271. Bonferroni-adjusted t-tests were used to investi-
gate at which presentation time levels categorization differed sig-
nificantly from identification performance. Figure 2 right panel
shows that categorization was better than identification at presen-
tation times longer than 36 ms.

These results suggest that detection, categorization, and identi-
fication are associated with different time courses. The observed
differences between the three recognition tasks in Experiment I,
however, might be owed to mechanisms other than visual process-
ing speed. In particular the lower performance in the categorization
task might be due to fading of categorization response represen-
tations in short term memory since participants answered the three
questions in the of detection, categorization, and identifica-
tion on 99.65% of the trials. Other, at least theoretically possible,
confounding factors that might also have biased the results of
Experiment I are backward priming (category response facilitation
due to the knowledge of the object’s identity), and experimental
design differences (there were two task relevant presentation in-
tervals for the detection task but only one for categorization and
identification task).

We tried to minimize memory, priming, and experimental de-
sign effects on recognition performance in Experiment II.

Experiment II: Single Response Control Experiment

Detection, categorization, and identification were measured in
separate experiments using a one-interval-forced-choice paradigm
(1IFC; Figure 3). In brief, every 2 s, participants saw one backward
masked image and had to indicate whether the shown image is of

a predefined category. In Experiment IIa ten naı̈ve participants
completed one detection and one categorization experiment to
compare detection and categorization performance; in Experiment
IIb another ten naı̈ve participants completed both one categoriza-
tion and one identification experiment to compare categorization
and identification performance. This experimental design mini-
mized memory and backward priming effects between different
recognition tasks by probing only one recognition task at a time.
Furthermore we eliminated design differences due to the employ-
ment of the same 1AFC task for all three recognition tasks.

Methods

Participants. Twenty naı̈ve participants (age range between
20 and 30 years; 12 females) participated in Experiment II (ten in
Experiment IIa and ten in Experiment IIb). All participants had a
normal or corrected-to-normal vision and were naı̈ve to the task
and stimuli. Each participant gave informed consent prior to the
experiment and was compensated 8€/hour for their participation.

Apparatus and stimuli. The apparatus and stimuli used in
Experiment IIa and IIb are identical to Experiment I.

Procedure. The same experimental procedure as outlined in
Figure 3 was used for Experiment IIa and IIb. In short participants
saw one image every 2 s and indicated for each image presentation
whether the shown image matched a predefined target. 50% of the
trials were target and 50% of the trials were non-target trials
(random assignment). Image presentation times in Experiment IIa
and IIb were chosen to highlight differences between the recogni-
tion tasks as suggested by Experiment I (see also Figure 2).
Consequently the image presentation times were 7, 14, 21, 28, and
78 ms in Experiment IIa and 21, 28, 36, 57, and 121 ms in
Experiment IIb. The noise presentation period (a new noise patch
was used on every trial) served also as response period. Partici-
pants’ task was to answer as quickly and as accurately as possible
whether the shown image matched the predefined target by press-
ing the target key (“z” or “/”) with their dominant hand and the

Figure 3. Experimental procedure used in Experiment II and III. The
presentation times for each screen are given in ms next to corresponding
screen. The keyboard key symbol indicates the screen that required par-
ticipants’ keyboard input, i.e. to start a run (see Methods section for the
definition of a run) and to give an answer. The target was defined only at
the beginning of a run. For sake of clarity the figure only shows the
presentation times for a categorization task in Experiment IIa along with
the correct key presses for a right handed participant.

41TIME COURSE AND SENSITIVITY OF VISUAL RECOGNITION

T
hi

s
do

cu
m

en
t i

s
co

py
ri

gh
te

d
by

th
e

A
m

er
ic

an
P

sy
ch

ol
og

ic
al

A
ss

oc
ia

tio
n

or
o

ne
o

f i
ts

a
lli

ed
p

ub
lis

he
rs

.
T

hi
s

ar
tic

le
is

in
te

nd
ed

s
ol

el
y

fo
r t

he
p

er
so

na
l u

se
o

f t
he

in
di

vi
du

al
u

se
r a

nd
is

n
ot

to
b

e
di

ss
em

in
at

ed
b

ro
ad

ly
.

non-target key with the non-dominant hand. 100 image presenta-
tions constituted a run. An experiment, e.g. detection experiment,
consisted of three of these runs (Experiment IIa consisted of one
detection and one categorization experiment; Experiment IIb con-
sisted of one categorization and one identification experiment). A
different target (e.g. bird in the categorization task or pigeon in the
identification task) was used for each run except for the detection
task, where the target was always an object image and the non-
target was always a scrambled object image (see Experiment I for
a description of the scrambling method). Participants were verbally
informed about the target by the experimenter prior to the run and
visually reminded of the target at the beginning of the run by the
presentation of the target word in the middle of the screen (“ob-
ject?” in the detection task, e.g. “bird?” in the categorization task,
e.g. “pigeon?” in the identification task). Non-targets were images
of other categories and non-exemplar images in the categorization
and identification task, respectively. Non-targets in the detection
task were scrambled object images. Recognition task testing
was counterbalanced across participants within Experiment IIa and
IIb. Note that recognition task was a within-subject-factor and
mean RT were calculated from hit trials only.

Results and Discussion

We compared detection, categorization, and identification per-
formance by means of d� and reaction times (RT). D� measures are
preferable to correction-for-guessing accuracy scores as they more
adequately assess the effect of partial knowledge on task perfor-
mance (Macmillan & Creelman, 2005). As for the d� calculation

we counted a correct target recognition as a hit and the recognition
of a non-target as a target as a false alarm.

Experiment IIA: Detection vs. Categorization

D� values as a function of presentation time are shown for the
detection and categorization task in Figure 4A top-left. Detection
d� values seem to be consistently higher than categorization d�
values. A repeated-measures ANOVA with presentation time and
recognition task as within-subject factors and d� as dependent
variable was used to investigate significant differences in detection
and categorization performance. We found significant main effects
of presentation time and recognition task, F(4, 36) � 73.14; p �
.001, �partial

2 � 0.890, and F(1, 9) � 44.93; p � .001, �partial
2 �

0.831 respectively. The interaction of presentation time and rec-
ognition task was also significant, F(4, 36) � 13.31, p � .001,
�partial

2 � 0.597 suggesting that performance differences between
the two recognition tasks depended on presentation time. Figure
4A bottom-left panel shows that detection performance was better
than categorization performance at all presentation time levels
except for 78 ms.

The mean reaction times of the detection and categorization task
are shown in the top right panel of Figure 4A. Reaction times are
longer for the categorization task than for the detection task at all
presentation time levels. We compared detection and categoriza-
tion reaction times of the hit trials in a repeated-measures ANOVA
with presentation time and recognition task as within-subject fac-
tors and mean RT as dependent variable. We found a significant
main effect of presentation time, F(4, 36) � 59.54, p � .001,

Figure 4. Results of Experiment II and III. Panel A, B, and C show the results of Experiment IIa (detection-
categorization) and Experiment IIb (categorization-identification), and Experiment III (detection-categorization),
respectively. All three panels are organized in the same way. The top-left panel shows d� as a function of
presentation time for each recognition task separately. The bottom left panel shows mean d� differences between
the two recognition tasks for each presentation time separately. The top-right panel plots mean RT as a function
of presentation time for each recognition task separately. The bottom right panel shows the mean RT differences
between the two recognition tasks for each presentation time separately. The letters SE or CI in top-right corner
of each graph indicate whether the bars indicate one standard error from the mean (SE) or the 95% Bonferroni
corrected confidence interval (CI).

42 DE LA ROSA, CHOUDHERY, AND CHATZIASTROS

T
hi

s
do

cu
m

en
t i

s
co

py
ri

gh
te

d
by

th
e

A
m

er
ic

an
P

sy
ch

ol
og

ic
al

A
ss

oc
ia

tio
n

or
o

ne
o

f …