Cognition 158 (2017) 33–43

Contents lists available at ScienceDirect

Cognition
journal homepage: www.elsevier.com/locate/COGNIT

Original Articles

From face processing to face recognition: Comparing three different

processing levels
G. Besson a,b,c,⇑,1, G. Barragan-Jason a,b,2, S.J. Thorpe a,b, M. Fabre-Thorpe a,b, S. Puma a,b,3, M. Ceccaldi c,d,
E.J. Barbeau a,b
a
Centre de Recherche Cerveau et Cognition, UPS, Université de Toulouse, 118 Route de Narbonne, 31062 Toulouse, France
b
CerCo CNRS UMR 5549, Pavillon Baudot CHU Purpan, BP 25202, 31052 Toulouse Cedex, France
c
Institut de neurosciences des systèmes, INSERM UMR 1106, Aix-Marseille Université, Faculté de Médecine, 27, Boulevard Jean Moulin, 13005 Marseille, France
d
CHU La Timone, Assistance Publique Hôpitaux de Marseille, 264 Rue Saint-Pierre, 13385 Marseille Cedex 5, France

a r t i c l e i n f o a b s t r a c t

Article history: Verifying that a face is from a target person (e.g. finding someone in the crowd) is a critical ability of the
Received 12 February 2016 human face processing system. Yet how fast this can be performed is unknown. The ‘entry-level shift due
Revised 26 September 2016 to expertise’ hypothesis suggests that - since humans are face experts - processing faces should be as fast
Accepted 6 October 2016
– or even faster – at the individual than at superordinate levels. In contrast, the ‘superordinate advantage’
hypothesis suggests that faces are processed from coarse to fine, so that the opposite pattern should be
observed. To clarify this debate, three different face processing levels were compared: (1) a superordinate
Keywords:
face categorization level (i.e. detecting human faces among animal faces), (2) a face familiarity level
Face recognition
Face categorization
(i.e. recognizing famous faces among unfamiliar ones) and (3) verifying that a face is from a target person,
Face detection our condition of interest. The minimal speed at which faces can be categorized (260 ms) or recognized
Familiarity as familiar (360 ms) has largely been documented in previous studies, and thus provides boundaries to
Reaction times compare our condition of interest to. Twenty-seven participants were included. The recent Speed and
Inversion effect Accuracy Boosting procedure paradigm (SAB) was used since it constrains participants to use their
fastest strategy. Stimuli were presented either upright or inverted. Results revealed that verifying that
a face is from a target person (minimal RT at 260 ms) was remarkably fast but longer than the face
categorization level (240 ms) and was more sensitive to face inversion. In contrast, it was much faster
than recognizing a face as familiar (380 ms), a level severely affected by face inversion. Face recognition
corresponding to finding a specific person in a crowd thus appears achievable in only a quarter of a
second. In favor of the ‘superordinate advantage’ hypothesis or coarse-to-fine account of the face
visual hierarchy, these results suggest a graded engagement of the face processing system across
processing levels as reflected by the face inversion effects. Furthermore, they underline how verifying
that a face is from a target person and detecting a face as familiar – both often referred to as ‘‘Face
Recognition” – in fact differs.
Ó 2016 Elsevier B.V. All rights reserved.

Abbreviations: RT, Reaction Time; minRT, minimal RT; HFC, Human Face 1. Introduction
Categorization; IFR, Individual Face Recognition; FFR, Familiar Face Recognition;
SAB, Speed and Accuracy Boosting procedure. Individuals are mostly recognized by their faces. Something we
⇑ Corresponding author at: Cyclotron Research Center, University of Liege, Allée
do daily for example is verifying that a face is from a target person
du 6 Août, B30, Liège 4000, Belgium.
(e.g. finding someone in the crowd). Classically, it is investigated
E-mail addresses: [email protected] (G. Besson), [email protected]
(G. Barragan-Jason), [email protected] (S.J. Thorpe), michele.fabre- in experimental tasks by proposing a verbal label to participants
[email protected] (M. Fabre-Thorpe), [email protected] (S. Puma), (e.g. ‘‘Brad Pitt”) and asking them whether subsequently presented
[email protected] (M. Ceccaldi), [email protected]. faces match or not with the label.
fr (E.J. Barbeau). Objects are usually categorized faster at the basic-level (e.g.
1
Present address: Cyclotron Research Center, University of Liège, Belgium.
2 bird vs. other animals) than at the superordinate (e.g. animal vs.
Present address: Institute for Advanced Study in Toulouse, France.
3
Present address: CLLE-LTC, Université de Toulouse, UT2J, France. vehicle) or subordinate-level (e.g. Indigo Bunting vs. other birds)

http://dx.doi.org/10.1016/j.cognition.2016.10.004
0010-0277/Ó 2016 Elsevier B.V. All rights reserved.
34 G. Besson et al. / Cognition 158 (2017) 33–43

(Rosch et al., 1976). The basic level is thus thought to be the entry performed at about 360 ms at the fastest (Barragan-Jason,
level at which people first process objects (Anaki & Bentin, 2009; Besson, Ceccaldi, & Barbeau, 2013; Barragan-Jason et al., 2012;
Johnson & Mervis, 1997; Jolicoeur, Gluck, & Kosslyn, 1984; Rosch Besson, Ceccaldi, Didic, & Barbeau, 2012), a quite long delay com-
et al., 1976; Tanaka, 2001; Tanaka & Taylor, 1991; Wong & pared to face categorization tasks.
Gauthier, 2007). However, this entry level may shift to the subor- What about verifying that a face is from a target person, the
dinate level with atypicality (e.g. penguins categorized faster than other face processing level aforementioned (Fig. 1A)? Such task
as birds; Jolicoeur et al., 1984) or with expertise (e.g. Indigo Bunt- has never been studied using a minimal RT approach. Under the
ing categorized as fast as birds by expert bird watchers; Johnson & entry level shift due to expertise hypothesis, such task should be
Mervis, 1997; Tanaka & Taylor, 1991). Humans are usually consid- performed faster – or at least as fast – than a superordinate level
ered to be face experts (Carey & Diamond, 1977; Carey, Schonen, & task (Anaki & Bentin, 2009; Tanaka, 2001). Under the superordi-
Ellis, 1992; Tanaka & Gauthier, 1997). Consistent with this idea, it nate advantage level hypothesis in contrast, such task should need
has been shown that faces are categorized as fast - or even faster more processing time than a superordinate task. In fact, some stud-
(see Anaki & Bentin, 2009) - at the individual level (e.g., as Brad ies have reported strikingly fast RTs (about 250 ms) in similar
Pitt) than at a superordinate level (e.g., as a human face) (Anaki tasks, suggesting it is worth investigating this issue in detail
& Bentin, 2009; Tanaka, 2001). (Lewis & Ellis, 2000).
However, at odds with such interpretation, neurophysiological How fast verifying that a face is from a target person would be
or neuroimaging studies have suggested that superordinate, relatively to detecting a face as familiar when no clue is available
coarse, information is processed before the more detailed infor- also remains unclear. In fact, the numerous terms used to refer
mation required for higher-level categorization (Large, Kiss, & to the verification that a face is from a target person (‘category-
McMullen, 2004; Löw et al., 2003; Martinovic, Gruber, & Müller, verification task’, Tanaka, 2001; ‘individual-level verification task’,
2008; Sugase, Yamane, Ueno, & Kawano, 1999; for faces, see Anaki & Bentin, 2009; or ‘face-identification task’, e.g. Delorme &
Goffaux et al., 2011). Interestingly, behavioral tasks also argue in Thorpe, 2001; Reddy, Reddy, & Koch, 2006) highlights how much
favor of such a coarse-to-fine access to perceptual representations its underlying mechanisms remain poorly understood. Specifically,
(Fabre-Thorpe, 2011; Hochstein & Ahissar, 2002), when studying does such a task need to rely on a person identity-level - a higher,
minimal reaction times (minRT) - i.e. the minimal processing time amodal and semantic level of representation, which would follow
necessary to give reliable responses (Rousselet, Macé, & Fabre- visual processes (Bruce & Young, 1986)? If so, verifying a face iden-
Thorpe, 2003). Aforementioned behavioral studies indeed classi- tity would be best described as a ‘face-identification task’ and
cally studied mean or median RTs without speed constraints. would be rather long, for instance close to familiarity tasks
However, these RTs could reflect processes which are not strictly (Valentine, 2001). In contrast, verifying a face identity could rely
necessary, such as verification or access to lexical information. For on facial diagnostic clues (e.g. specific facial features characteristic
example, access to basic words could be shorter than access to of a face) that could help preparing and optimizing visual process-
superordinate words since they are more frequently used. Using ing through top-down strategies, such as preactivation and atten-
the minRT approach, Macé, Joubert, Nespoulous, and Fabre- tional selection (e.g. Eimer, 2014). In this case, it could be quite
Thorpe (2009) showed a superordinate advantage compared to fast, and close to categorization tasks, which rely on similar
the basic level when animals had to be categorized, a finding fur- mechanisms.
ther confirmed in other studies (Kadar & Ben-Shahar, 2012; In this study, we compared performance speed in an Individ-
Loschky & Larson, 2010; Prab, Grimsen, König, & Fahle, 2013; ual Face Recognition task (i.e. verifying that a face is from a tar-
Vanmarcke & Wagemans, 2015; Vanmarcke et al., 2016). Such get person) to a Human Face Categorization task and to a
superordinate-level advantage was shown to be independent of Familiar Face Recognition task. The difference between these
stimuli duration or target and distractor diversity (Poncet & conditions is visually schematized in Fig. 1A. Interestingly, the
Fabre-Thorpe, 2014). distinction between Individual Face Recognition and Familiar
The prediction of the superordinate advantage level hypothesis Face Recognition conditions is not always clear in the literature
for faces would be that faces would be categorized faster at the whereas they may rely on different processes and hence yield
superordinate than at the individual level, despite the expertise different RTs.
advantage. To date, only one study compared different levels of As already presented, the speed at which faces can be processed
face categorization using minimal RTs. In this study, participants is largely known for either Human Face Categorization (minRT:
had to perform a ‘human face vs. animal face’ superordinate cat- 260 ms) or Familiar Face Recognition (minRT: 360 ms). The
egorization task, which was contrasted with a ‘familiar face vs. aim of this study is to assess the speed of Individual Face Recogni-
unfamiliar face’ subordinate recognition task. Results were clear tion compared to these boundaries (Fig. 1B), and thus to determine
as the superordinate task was performed much faster (minRT: what temporal hierarchy, if any, there is between these three levels
250 ms) than the subordinate (440 ms) (Barragan-Jason, of face processing. To test the entry level shift related to expertise
Lachat, & Barbeau, 2012). Although particularly strong, such an or the superordinate level hypotheses, we will compare minimal
effect was expected since the superordinate categorization task RTs in the Individual Face Recognition condition to the Human
can rely on the detection of low-level features (Crouzet, Face Categorization and Familiar Face Recognition conditions
Kirchner, & Thorpe, 2010; Rossion & Caharel, 2011; Rossion & (Fig. 1C). For such comparisons to make sense, it is necessary to
Jacques, 2011) and hence be very fast (about 260 ms, reviewed constrain participants to use their fastest strategy in each condi-
in Fabre-Thorpe, 2011). In contrast, participants had to recognize tion (Barragan-Jason et al., 2013). We thus used the Speed and
famous faces among unknown ones in the subordinate (i.e. famil- Accuracy Boosting procedure (SAB), a recent procedure based on
iarity) task. They did not know in advance which famous faces a go/no-go paradigm with a response deadline (Besson et al.,
would be presented. Each face thus had to be processed up to 2012) in which responses must be provided before a constraining
the individual level in a bottom-up fashion before a familiarity time limit, set in this study at 600 ms (Fig. 1D). Last, since face
signal could be triggered. Such level of processing thus refers to inversion is known to disrupt holistic processing and access to face
a particular kind of face recognition task, for which no clue is configuration, we investigated the effect of face inversion on these
available before the face is processed, and more akin to unexpect- different conditions, by also running all three with inverted stimuli
edly meeting an acquaintance in the street (Fig. 1A). Several stud- (Farah, Wilson, Drain, & Tanaka, 1998; Maurer, Grand, & Mondloch,
ies have now reported that such face recognition task can be 2002; Rossion, 2008; Yin, 1969).
 G. Besson et al. / Cognition 158 (2017) 33–43 35

A Finding a human face Finding the person one is Unexpectedly meeting

among other objects… looking for in a crowd... an acquaintance in the
street…

HFC IFR FFR Hey!

Is there ? Where is ? ?!
anybody?…
vs he? …
vs You? Here!

Human Face Individual Face Familiar Face

Categorization Recognition Recognition

B HFC FFR
~260 ms* ~380 ms*
?

IFR *e.g. Barragan-Jason et al., 2012

? ?

C
Human faces
Task: HFC vs animal faces

"Sarkozy’s" faces
Task: IFR vs paired unknown faces

Famous faces
Task: FFR vs unknown faces

D
• Target: Hit
+
• Distractor: FA
« The SAB procedure »
Go!
Next Trial

600
RT
Deadline

+ 100
0
Onset

No-Go (no RT)

Fixation
cross

• Target: Miss
• Distractor: CR

Fig. 1. Experimental design. (A) Schema depicting different levels of face processing: Human Face Categorization (HFC), Individual Face Recognition (IFR), Familiar Face
Recognition (FFR). (B) If the minimal processing time needed for Human Face Categorization and Familiar Face Recognition can be predicted from previous studies, different
hypotheses are possible regarding the minimal processing time needed for Individual Face Recognition. (C) Samples of the stimuli used in each of the three tasks. (D) SAB
procedure. See Section 2.4 for details.

2. Material and methods blocks of 140 stimuli: one using upright stimuli, the other using
inverted stimuli. No stimulus was repeated, even inverted. The
2.1. Participants order of the three tasks was pseudo-randomized and counter-
balanced across participants, as well as the order of upright and
27 participants (15 women) were included in this study (med- inverted conditions.
ian age: 24 [21–27], 2 left-handed). All participants signed
informed consent and had normal or corrected-to-normal vision. 2.3. Tasks

2.2. Experiment In the Familiar Face Recognition task, targets were famous faces
of different persons (persons were never repeated) and distractors
The experiment included 3 tasks: (1) a Familiar Face Recogni- unknown faces. In the Individual Face Recognition task, targets
tion task, (2) an Individual Face Recognition task and (3) a Human were different picture of the same famous person and had to be
Face Categorization task. Each task was made of 2 consecutive recognized among matched unknown faces (distractors). One block
36 G. Besson et al. / Cognition 158 (2017) 33–43

(the upright or the inverted one) consisted in recognizing faces of Unknown human faces and animals faces of the Human Face
Nicolas Sarkozy (NS), President of France during the period the Categorization task were randomly chosen from a previous similar
experiment was run; the other block consisted in recognizing faces experiment (Barragan-Jason et al., 2012).
of Johnny Hallyday (JH), a highly famous French rock singer. Hence, All pictures for the three experiments were in grayscale (256
the upright block was performed with one of the famous person levels). Each picture was framed manually around the face follow-
target, whereas the inverted one was performed with the other ing the same procedure (i.e. a rectangle delimited at the bottom by
famous person target, a choice that was pseudo-randomized and the chin, on the side by the point between the face itself and the
balanced across participants. In the Human Face Categorization ear or, when the ear was masked in three-quarter profiles, the far-
task, targets were human faces and distractors animals faces all thest point between the eyebrow arch and the cheekbone, and at
presented randomly. Each block was made of 70 targets intermixed the top by the midpoint of the front hairline, extrapolated if neces-
with 70 distractors. sary). Pictures could then be resized so that each face had the same
size and could be cropped all identically around the face, using a
2.4. Procedure homemade script on Matlab. Thus, all pictures were similar
close-up grayscale pictures of centered faces of the same size
Each block was run using the Speed and Accuracy Boosting (208  279 pixels, visual angle: 4.7  6.3°). Lastly, all stimuli
procedure (SAB) with a response deadline set at 600 ms from were equalized to the same luminance (mean grey-level: 108.7)
stimulus onset, a deadline inferred from previous studies and contrast (computed as the standard error of pixels luminance:
(Besson et al., 2012; Besson et al., 2015). Inspired by different 54) across conditions.
approaches (the Speed-Accuracy Trade-off procedure, e.g.
Dosher, 1976; response-deadline procedures, e.g. Reed, 1973 or
more recently Bowles et al., 2007; minimal reaction times, e.g. 2.6. Set-up
Rousselet et al., 2003), the SAB procedure specifically aims at
studying the reaction times distribution while constraining sub- Participants sat in a dimly lit room, at 90 cm from a computer
jects to use their fastest strategy (for a discussion, see Besson screen piloted by a PC. Image presentation and behavioral
et al., 2012). Briefly, it is based on a classical Go/No-Go task, con- responses recordings were carried out using the E-prime v2. Partic-
strains participants to answer before a response deadline (to ipants responded to the stimuli by raising their fingers from a
boost speed), and provides an audio-feedback - positive if the custom-made infrared response pad.
item was a target (hit), negative if the item was a distractor
(false-alarm) - at the response (to boost accuracy). If no response 2.7. Minimal reaction times
is made before the response deadline, the response is considered
as a No-Go-response and, at the response deadline, an audio- To obtain an estimation of the minimal processing time
feedback is played - positive if the item was a distractor (correct required to recognize targets, the minimal behavioral reaction
rejection), negative if the item was a target (miss) (Fig. 1D). time (minRT) was computed by determining the latency at
Before each item presentation, a fixation cross is displayed for a which correct go-responses (hits) started to significantly out-
pseudo-random time between 300 and 600 ms. Items are pre- number incorrect go-responses (false-alarms) (Rousselet et al.,
sented, one by one, in the center of a grey screen, for 100 ms 2003). For each task, analyses were performed both across trials
(comprised in the response deadline). The SAB is highly demand- (by pooling together all trials from all participants for a given
ing and training is necessary. Each block was preceded by a train- condition) and across participants. Across trials analyses have
ing block (20 targets to be recognized among 20 distractors), been used in previous studies (Barragan-Jason et al., 2012;
which could be repeated if needed following participants’ request. Besson et al., 2012; Rousselet et al., 2003) and are like building
Training stimuli were not re-used in any experimental block. A a ‘‘meta-participant”, reflecting the performance over all the
self-paced pause was proposed each 20 trials. population. MinRTs across trials were computed using 10 ms
time bins and determined as the middle of the first bin that
2.5. Stimuli was significant, v2-test, p < 0.05, followed by at least three sig-
nificant consecutive bins. Across participants, in order to accom-
Sample stimuli for each task are presented in Fig. 1C. Famous modate for the lower statistical power than across trials data
faces in the Familiar Face Recognition task were selected as the since there were fewer trials, we used 40 ms time bins and a
best recognized famous faces of a large database used in previous Fisher’s exact test (p < 0.05). A minRT can’t be computed if the
experiments with participants of the same age as in this study distribution of hits and false alarms are too close. Thus some
(Barragan-Jason et al., 2013). Distractors for famous faces were participants don’t have a minRT, in particular when d’ are low
unknown faces randomly selected from the same study. Unknown (details reported in Section 3).
faces were chosen so that they ‘‘looked like” they could be famous.
Pictures for the Individual Face Recognition task were selected
from the web. All original pictures presented a face area of at least 2.8. Statistical analyses
200  200 pixels. No blurry picture was included. We avoided as
much as possible pictures with lots of details in the background Performance (accuracy) and bias were computed using d’
and with any objects hiding the face (e.g. hand, microphone, other and C based on the signal detection theory (corrected accord-
person, etc.). All pictures were chosen relatively close to one proto- ing to Snodgrass & Corwin, 1988). Participants’ success on a
type face (e.g. in the same period of life, color of hair, etc.), but we task was determined statistically (v2-test between hits and
avoided choosing pictures too similar (e.g. from the same photo false alarms (FA) among targets and distractors, p < 0.05). As
shooting, etc.). The distractor sets were made by matching one- parametrical conditions were largely met (normality checked
by-one an unknown face picture with each target picture with Lilliefors test, p < 0.05; variance equality tested with
(Fig. 1C). Pictures were as much as possible matched on the types Brown & Forsythe test, p < 0.05), statistical comparisons were
of clothes, type and color of hair (and beard), color of eyes, age, computed using ANOVAs. For RTs, mean and standard devia-
paraphernalia or hiding object if needed, head orientation, face tion were computed based on formula defined for lognormal
expression (e.g. smiling, laughing, neutral, talking, etc.). distribution.
 G. Besson et al. / Cognition 158 (2017) 33–43 37

3. Results Categorization (see Fig. 2A). The Familiar Face Recognition was
much more difficult than the Individual Face Recognition and
3.1. Across participants accuracy Human Face Categorization and not every participant succeeded
on the task. In the upright condition, three participants did not suc-
A repeated measures two-way ANOVA on accuracy with task and ceed on the Familiar Face Recognition and were thus discarded from
orientation as factors revealed a clear main effect of the task the study, while in the inverted condition only a few succeeded at
(F(2, 22) = 784.6; p < 0.0005) and of the orientation (F(1, 23) this condition (see Table 1). Furthermore, the effect of inversion
= 402.3; p < 0.0005), as well as a significant interaction between was computed (i.e. the difference between inverted face accuracy
them (F(2, 22) = 60.2; p < 0.0005). Accuracy was smaller in the and upright faces accuracy divided by the upright faces accuracy;
Familiar Face Recognition condition than in the Individual Face e.g. Russell, Duchaine, & Nakayama, 2009) and showed a significant
Recognition, which itself was smaller than in the Human Face difference between the three task (F(2, 22) = 151.4; p < 0.0005).

Fig. 2. Across participants analyses. (A) Comparison of performances, bias and minRTs computed across participant in each task. The main condition of interest, the Individual
Face Recognition, is depicted in yellow. (B) Comparison of the effects of inversion on accuracy between tasks. * p < 0.05, *** p < 0.001, n.s.: p > 0.05. (minRTs: p-values were
corrected with Bonferroni correction for multiple comparisons). Each circle is a participant. FFR = Familiar Face Recognition, HFC = Human Face Categorization,
IFR = Individual Face Recognition. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 1
Performance on the three tasks for upright and inverted faces (N = 27). FFR = Familiar Face Recognition, HFC = Human Face Categorization, IFR = Individual Face Recognition,
N = number of participants, RT = Reaction Time, SD = standard deviation.

Upright FFR Inverted FFR Upright IFR Inverted IFR Upright HFC Inverted HFC
task task task task task task
Accuracy (d’) Mean 1.06 0.13 3.96 2.18 4.20 3.85
SD 0.41 0.28 0.47 0.37 0.52 0.56
Range [0.50; 2.16] [ 0.32; 0.71] [2.64; 4.91] [1.44; 2.98] [2.78; 4.91] [2.47; 4.91]
Across trials 1.00 0.12 3.98 2.17 4.25 3.84
Bias (C) Mean 0.19 0.26 0.19 0.16 0.14 0.18
SD 0.37 0.70 0.27 0.17 0.23 0.18
Range [ 0.37; 1.13] [ 0.62; 2.46] [ 0.73; 0.32] [ 0.42; 0.16] [ 0.56; 0.32] [ 0.60; 0.19]
Across trials 0.18 0.18 0.26 0.18 0.19 0.18
Hits (%) Mean 63 45 99 90 99 98
SD 16 20 1 3 2 3
Range [27; 84] [0; 73] [94; 100] [83; 94] [90; 100] [86; 100]
Across trials 63 45 99 90 99 98
FAs (%) Mean 25 41 4 18 3 4
SD 12 22 4 8 2 3
Range [4; 49] [0; 74] [0; 16] [4; 34] [0; 9] [0; 10]
Across trials 63 45 99 90 99 98
Hits median RTs Mean 452.9 426.0 352.5 389.8 323.7 336.1
SD 31.7 95.1 25.5 20.4 26.8 27.7
Range [392; 548] [221; 544.5] [318; 418] [355.5; 427] [289.5; 396] [295; 397.5]
Across trials 449 423.5 351 388 320 334
Obtained a minRT/succeeded the task N (%) 22/24 (91.67%) 2/4 (50.00%) 24/24 24/24 24/24 24/24
Minimal RTs Mean 443.9 441.8 301.7 351.7 273.4 283.4
SD 66.0 57.2 26.8 27.3 25.6 25.6
Range [320; 560] [400; 480] [240; 360] [280; 400] [240; 320] [240; 360]
Across trials 380 – 260 290 240 240
38 G. Besson et al. / Cognition 158 (2017) 33–43

Post-hoc analyses showed that this effect was larger in the Familiar condition, minRTs were longer in the Familiar Face Recognition
Face Recognition than in the Individual Face Recognition (Familiar than in the Individual Face Recognition (difference across partici-
Face Recognition: 88.1%, SD = 29.6%; Individual Face Recognition: pants: 142 ms, SD = 67 ms), itself longer than Human Face Catego-
44.5%, SD = 9.9%; p < 0.0005), which itself was larger than in the rization (28 ms, SD = 32) (see also Fig. 2A, Table 1). Second, a
Human Face Categorization (8.1%, SD = 10.7%; p < 0.0005; Fig. 2B). repeated measures two-way ANOVA with task and orientation as
factors across Individual Face Recognition and Human Face Catego-
3.2. Across participants minRTs rization tasks revealed a main effect of the task (F(1, 23) = 82.4;
p < 0.0005) and of the orientation (F(1, 23) = 56.1; p < 0.0005), and
MinRTs could not be calculated for all participants in the Famil- a significant interaction between them could be observed (F
iar Face Recognition, in particular in the inverted condition. Hence, (1, 23) = 28.8; p < 0.0005) (see also Fig. 2A, Table 1).
the 2  3 repeated measures two-way ANOVA design with task
and orientation as factors were not carried out. Nevertheless, an 3.3. Across trials analyses
unbalanced one-way ANOVA across upright conditions on log-
transformed minRTs first revealed a significant effect of the task Because minRTs by nature is sensitive to the lack of statistical
on minRTs (F(2, 69) = 115.2; p < 0.0005). Indeed, in the upright power, analyses across trials were conducted (Fig. 3) using both

Fig. 3. Performances across time. (A) Performances obtained on each of the three tasks. Across trials distributions of RTs obtained in each task in the upright or inverted
condition. Vertical bars and numbers correspond to across trials minRTs obtained in each task. (B) Cumulated d’ for the three tasks with standard error of the mean computed
for each 1-ms time step. FFR = Familiar Face Recognition, HFC = Human Face Categorization, IFR = Individual Face Recognition.
 G. Besson et al. / Cognition 158 (2017) 33–43 39

distributions of RTs and cumulated d-prime. Such analyses are (F(2, 38) = 32.1; p < 0.0005), implying that, in each condition, the
similar to the results of a meta-participant and provide a summary speed did not depend upon detection rate alone. Post-hoc analyses
of the results. on this selection of targets showed that median RTs were longer in
the Familiar Face Recognition than in the Individual Face Recogni-
3.4. Bias tion (Familiar Face Recognition: 446 ms, SD = 41 ms; Individual
Face Recognition: 369 ms, SD = 41 ms; p < 0.0005), which them-
A repeated measures two-way ANOVA on bias with task and selves were longer than in the Human Face Categorization
orientation as factors revealed a clear main effect of the task (F (326 ms, SD = 18 ms; p < 0.0005). Similar results were observed
(2, 22) = 8.2; p = 0.002), but not of the orientation (F(1, 23) = 0.2; on distractors.
p = 0.68), and no interaction between them (F(2, 22) = 0.69;
p = 0.51). For upright, as for inverted stimuli, a conservative bias
3.6. Supplementary analyses
was observed in the Familiar Face Recognition condition, contrast-
ing with a liberal and identical bias observed in the Individual Face
3.6.1. Effect of target identity on Individual Face Recognition
Recognition and Human Face Categorization (Fig. 2A, Table 1).
performance
Results of Individual Face Recognition tasks are presented in
3.5. Across-targets analyses Table 2. Overall, no statistical difference was observed between
the two targets (Nicolas Sarkozy, IFR-NS or Johnny Hallyday, IFR-
In order to investigate any effect of individual stimulus, for each JH) on any variable (accuracy, bias or minRT). Nonetheless, it
target, a detection rate (i.e. percentage of go-responses) and a med- should be noted that there was a trend towards an effect of the tar-
ian RT were computed (Fig. 4). Unbalanced one-way ANOVAs get in the two-way ANOVA on accuracy (F(1, 44) = 3.73, p = 0.06),
across upright conditions, revealed a main effect of the task for which could be explained by a main simple effect of the target in
both variables (detection rate, F(2417) = 463; p < 0.0005; median the inverted condition (F(1, 50) = 6.47, p = 0.015).
RT, F(2417) = 669; p < 0.0005). When focusing on very well detected
targets across the three conditions (i.e. targets detected between
90% and 95% of time, i.e. very well detected targets excluding those 3.6.2. Effect of Individual Face Recognition preparation on the first
at ceiling, and with detection rates not different among conditions, target
F(2, 38) = 0.75; p = 0.5), a main effect on median RTs still remained RTs of each participant were ordered by the rank of presenta-
tion of the targets, with no RT if no go-response was made on a
particular target by a given participant (omission). We focused
100 on the RTs of the first target (targets were different faces across
participants as targets were presented randomly for each partici-
(% go-responses)

80
Detection rate

pant) (Fig. 5A). Importantly, the large difference observed between

60 Familiar Face Recognition and Individual Face Recognition
reported in earlier results across all targets was apparent right
40 from the beginning, i.e. for the first target.
HFC In a complementary analysis, we focused on the first go-
20 IFR responses (rather than on the first target as in the previous analy-
FFR
sis). Each first go-response given either on a target or a distractor
0
0 300 400 500 600
was thus correct (hit) or incorrect (false-alarm). Therefore, this
allowed us to compute minRTs (Fig. 5B). Results were largely con-
Median RT
(on go-responses, ms) vergent with the previous analysis. No error was made on Human
Face Categorization upright and only one on Individual Face Recog-
Fig. 4. Across-targets analysis of detection rate as a function of median RT on nition upright. This contrasted with the Familiar Face Recognition
targets. Each circle is a target. for which results were much poorer.

Table 2
Performance on the two Individual Face Recognition (IFR) tasks. N = number of participants, RT = Reaction Time, SD = standard deviation.

Upright IFR-NS Inverted IFR-NS Upright IFR-JH Inverted IFR-JH

Underwent the task N 12 15 15 12
Accuracy (d’) Mean 4.00 2.34 3.89 1.93
SD 0.62 0.20 0.36 0.37
Range [2.64; 4.91] [2.04; 2.75] [2.98; 4.49] [1.44; 2.98]
Across trials 3.90 2.35 3.92 1.92
Bias (C) Mean 0.22 0.13 0.18 0.18
SD 0.31 0.19 0.26 0.19
Range [ 0.73; 0.32] [ 0.42; 0.25] [ 0.52; 0.32] [ 0.41; 0.16]
Across trials 0.29 0.14 0.19 0.20
Hits median RTs Mean 347.2 392.0 356.8 389.3
SD 20.7 23.5 27.2 14.8
Range [318; 400.5] [355.5; 427] [318; 418] [364; 409]
Across trials 347 390 354 388
Obtained a minRT/succeeded the task N 12/12 15/15 15/15 12/12
Minimal RTs Mean 290.1 346.8 309.4 356.7
SD 25.3 30.0 23.8 20.8
Range [240; 320] [280; 400] [280; 360] [320; 400]
Across trials 260 300 260 330
40 G. Besson et al. / Cognition 158 (2017) 33–43

A 600
B 600

500 500 517

462
Median RTs (ms)

463
400 392

RT (ms)
326 400
300 372

332
300 305
200
282
Human Face Categorization
100 Individual Face Recognition
200
Familiar Face Recognition
0
0 10 20 30 40 50 60 70
Targets ordered by rank of presentation
The first go-responses

Fig. 5. Supplementary analyses. (A) Comparison of median RTs obtained on targets ordered by rank of presentation. Colored rectangles represent significance between
conditions. (B) RTs obtained for the first go-responses, in each of task. Each dot represents an RT obtained by a participant either on a target (i.e. hit, in blue) or on a distractor
(i.e. false alarm, in red). Horizontal lines indicate the minRT. No minRT could be computed for Familiar Face Recognition because of too many false-alarms. FFR = Familiar Face
Recognition, HFC = Human Face Categorization, IFR = Individual Face Recognition, upr: Upright condition, inv = Inverted condition. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

3.6.3. Effect of practice trast, Individual Face Recognition and Familiar Face Recognition
We then focused on the RTs of the first target compared to the appeared very different. Individual Face Recognition was 120–
rest of the RTs in the same condition to investigate a possible role 140 ms faster than Familiar Face Recognition, with much higher
of practice or of fatigue. Statistical differences between each pair of accuracy and liberal vs conservative biases, suggesting that partic-
rank of targets presented were computed. No repetition effect was ipants used different strategies. This difference in delays, as the dif-
observed (Fig. 5A). ference in accuracy, held since the first response (Fig. 4) and was
observed even considering comparably well-detected targets
4. Discussion (Fig. 5). Hence, Individual Face Recognition appears highly similar
to Human Face Categorization, widely regarded as a ‘‘fast” catego-
The aim of this study was to assess the difference between three rization task (Fabre-Thorpe, 2011), but sharply different from
different levels of face processing: (1) face categorization (i.e. Familiar Face Recognition.
detecting human faces among animal faces) (Human Face Catego- Since Familiar Face Recognition was tested under conditions
rization condition), (2) face familiarity (i.e. recognizing famous that enforced reliance on familiarity rather than on identification
faces among unfamiliar ones) (Familiar Face Recognition condi- (for discussion see Besson et al., 2012 and Barragan-Jason et al.,
tion) and (3) verifying that a face belongs to a target person (Indi- 2013), it appears unlikely in Individual Face Recognition that a per-
vidual Face Recognition condition, Fig. 1A). Familiar Face son identity-level (amodal and semantic level of representation) is
Recognition and Individual Face Recognition conditions are com- reached in order to trigger such fast responses. In contrast, it is
monly referred to with the same name, i.e. ‘‘face recognition task”, more plausible that Individual Face Recognition is of the same nat-
and one further goal of this study was to assess whether they were ure than Human Face Categorization, albeit delayed. Indeed, Indi-
alike or whether they differentiated. Because the speed of process- vidual Face Recognition may preactivate a diagnostic visual
ing Human Face Categorization (minRT: 260 ms) and Familiar representation of the target face, as Human Face Categorization
Face Recognition (minRT: 360 ms) are largely documented, our may preactivate a diagnostic visual representation of human face-
main variable of interest was related to the Individual Face Recog- ness, that needs to be verified when the input stimulus is pro-
nition speed of processing and how it would compare to either cessed. Human Face Categorization is defined as a category-
Human Face Categorization or Familiar Face Recognition verification task. Likewise, Individual Face Recognition could be
(Fig. 1B). Results revealed that Individual Face Recognition is much best described as a category-verification task at a subordinate-
faster than Familiar Face Recognition, but, albeit longer than level of face processing (individual level). Such a description of
Human Face Categorization, still strikingly rapid: reliable Individual Face Recognition is not new (e.g. Tanaka, 2001; see also
responses can be given in a quarter of second only, strongly con- Mack & Palmeri, 2011; Palmeri & Gauthier, 2004) but not unani-
straining models of Individual Face Recognition processing. mous (e.g. Anaki & Bentin, 2009; Delorme & Thorpe, 2001; Reddy
et al., 2006). We claim that the present findings, obtained using
4.1. Differences between Individual Face Recognition and other face an assessment of minimal RTs, argue strongly in favor of such a
processing levels description. Interestingly, evidence show that tasks at the same
level than the Human Face Categorization or Individual Face
Individual Face Recognition took 20–30 ms longer than Recognition are achievable in the near-absence of attention
Human Face Categorization (across-participants mean minRTs: (Reddy et al., 2006; VanRullen, Reddy, & Koch, 2004), which argues
302 ms vs. 273 ms; across-trials minRTs: 260 ms vs. 240 ms). Hold- again for a similar mechanisms behind both these levels.
ing since the first response (66 or 50 ms difference depending on
the analysis), this short difference is unlikely to be related to learn- 4.2. Effects of face inversion
ing or repetition of similar stimuli. Importantly, Individual Face
Recognition and Human Face Categorization appeared strongly Different face inversion effects were observed on Individual
similar, as evidenced by similar high accuracies, similar conserva- Face Recognition and Human Face Categorization, as the effect of
tive bias, and similar RT distributions across trials (Fig. 3A). In con- inversion was clearly stronger in Individual Face Recognition than
 G. Besson et al. / Cognition 158 (2017) 33–43 41

in Human Face Categorization, both on accuracy (accuracy dropped Ahissar, 2002). Interestingly, more additional time was necessary
by 44.5% for Individual Face Recognition vs. 8.1% for Human Face from a superordinate to a basic level on animal categorization
Categorization) and minRTs (Fig. 2). Thus face inversion partly dis- (e.g. Macé et al., 2009), than here on face categorization (respec-
rupted Individual Face Recognition (while still allowing a fair level tively 40–65 ms vs. 20–30 ms), perhaps due to the expertise of
of performance). It also affected Human Face Categorization, but humans for human faces.
barely, consistently with a previous study that reported no inver- Such a result is at odds with studies run without speed con-
sion effect on d’ or minRTs in a Human Face Categorization task straints (Anaki & Bentin, 2009; Tanaka, 2001). A plausible interpre-
(Rousselet et al., 2003; of note, inverted stimuli were intermixed tation is that in studies with unconstrained time, processing may
with upright stimuli in this study). In contrast, face inversion com- spontaneously proceed until the a priori most relevant or most nat-
pletely impaired Familiar Face Recognition performance (accuracy ural entry-level, which in the case of faces is the identity level
dropped by 88.1%). Inversion is known to disrupt face recognition, (Bruce & Young, 1986; Valentine, 2001). Along the way, coarser
disproportionally compared to non-face objects (Brown, Huey, & levels and categorization may be processed, however implicitly.
Findlay, 1997; Valentine, 1988; Valentine & Bruce, 1988; Yin, This pattern suggests that once this spontaneous entry-level is
1969). This face inversion effect – a marker of the holistic process- reached, categorizing ‘back’ to a coarser level, involves further pro-
ing of faces (e.g. Farah et al., 1998; Maurer et al., 2002; Rossion, cessing. Hence, in studies with unconstrained time, RTs may be as
2008) – is one of the critical characteristics of the face processing long – or even longer – at the superordinate level than at the basic
system. Importantly, these inversion effects appear highly consis- or subordinate level (Anaki & Bentin, 2009; Tanaka, 2001). The
tent with the observed minRTs since Human Face Categorization inverse is observed in studies using time constraints as in the pre-
appears to require minimal or no holistic processing and is the sent study since participants are this time constrained to shift their
fastest, Individual Face Recognition appears to require partial natural entry-level to a more specific one. Our study show that par-
holistic processing and takes slightly longer, while Familiar Face ticipants can actually do it well, quickly and in a way that respects
Recognition requires complete holistic processing and takes much the visual hierarchy. This also suggests that experiments aiming at
longer (however other processes may also be involved in Familiar assessing the hierarchy of these levels should take care of the ten-
Face Recognition, e.g. bottom-up processing of the stimulus as a dency of participants to spontaneously reach the natural entry-
whole, etc.). level.

4.3. A hierarchy of face processing 4.4. The speed of Individual Face Recognition: implications

Our results overall lend strong support in favor of the superor- Assessing minimal processing time exerts strong constraints on
dinate advantage hypothesis, which holds that the superordinate the models of underlying neural mechanisms. Making a decision
level is processed first (Human Face Categorization) before any and a motor response would take 110 ms (Kalaska &
other subordinate level (Individual Face Recognition or Familiar Crammond, 1992; VanRullen & Thorpe, 2001). Hence, brain signals
Face Recognition) (Macé et al., 2009). Even if Individual Face triggering Human Face Categorization, Individual Face Recognition
Recognition is fast and appears similar to Human Face Categoriza- and Familiar Face Recognition responses could arise at post-
tion in comparison to Familiar Face Recognition, it has special char- stimulus onset latencies as short as 130 ms ,150 ms, and
acteristics such as a sensitivity to face inversion that makes it 270 ms respectively (i.e. 110 ms before the minimal time
slower. In this study, the hierarchy revealed between Human Face needed to perform the tasks, as reflected by across trials minRTs).
Categorization, Individual Face Recognition and Familiar Face Actually, the onset of Human Face Categorization brain signals
Recognition regarding the minimal processing time and the face (130 ms) appears very similar with the 130-ms latency at
inversion effect may therefore reflect the different extent to which which EEG first changes following a (perceived) shift of faces
face-specific processes are engaged. More precisely, detecting (Jacques & Rossion, 2006) or with the 125-ms latency at which
human-faceness in Human Face Categorization may barely require EEG during a similar Human Face Categorization task started to
any face specific processing; Individual Face Recognition may correlate with RTs (based on a Multi-Variate Pattern Analysis;
partly rely on some (relatively low-level) aspects of face configura- Cauchoix, Barragan-Jason, Serre, & Barbeau, 2014). Given how close
tion (for example on first-order relation, Maurer et al., 2002); Individual Face Recognition is to Human Face Categorization, it
Familiar Face Recognition would require the highest level of face thus appears likely that brain signals triggering Individual Face
configuration (for example holistic processing, Maurer et al., Recognition should be seen around 150 ms or even a bit earlier.
2002; Rossion, 2008). Of note, this hierarchy may only apply to This would correspond to the onset of the N170, a face-sensitive
stimuli easily recognizable. In more ambiguous or noisy situations component observed in evoked-related potentials, starting at
(e.g. cluttered or mooney faces), face categorization may not com- 130 ms and peaking around 160–170 ms. The N170 is thought
pletely rely on the detection of diagnostic low-level features any- to index access to face representation and to reflect the engage-
more (Crouzet et al., 2010; Rossion & Caharel, 2011; Rossion & ment of a face-specific processing system independently of task
Jacques, 2011), and may require more face-specific processes demand (Rossion & Jacques, 2011). This seems all the more plausi-
(e.g. holistic processing; e.g. Mack, Gauthier, Sadr, & Palmeri, ble since the Individual Face Recognition shows an inversion effect,
2008). Interestingly though, the face inversion effect we found in to which the N170 is highly sensitive. Of note, the 20–30 ms dif-
the Individual Face Recognition condition suggests that the face ference in speed between Human Face Categorization and Individ-
diagnostic visual representation preactivated by top-down prepa- ual Face Recognition does not necessarily imply a serial stage from
ration is not simply related to the preactivation of low-level visual Human Face Categorization to Individual Face Recognition. Both
features. Human Face Categorization and Individual Face Recognition may
Expanding previous behavioral studies on animal categorization share similar processes, for example an obligatory face detection
to face categorization, this result is consistent with the idea that a stage gating any face processing (e.g. Tsao & Livingstone, 2008),
finer face representation is needed in order to categorize a face at but since both may rely on the preactivation of a diagnostic visual
the individual level than as ‘‘human”. This appears in line with a representation through top-down mechanisms, processing at this
coarse-to-fine access to face representations as has been posited level of representation of the input stimulus may be optimized dif-
(Large et al., 2004; Löw et al., 2003; Martinovic et al., 2008; ferentially. Hence, it is possible that specific Individual Face Recog-
Sugase et al., 1999; see also Fabre-Thorpe, 2011; Hochstein & nition processes start even before 130 ms.
42 G. Besson et al. / Cognition 158 (2017) 33–43

Thus, the prediction can be made that Individual Face Recogni- findings are unlikely to be contaminated by a simple effect of our
tion relies on specific neural activity developing as early as 150 ms, stimuli. First, RTs across conditions were sensibly different even
or possibly earlier and consequently that the brain regions for targets with comparable level of detection (Fig. 4). In the same
involved in this activity are those involved in early face processing, vein, no correlation of median RTs and accuracy was observed
i.e. posterior areas of the visual ventral stream. Top-down factors, across targets in the Familiar Face Recognition, which would have
such as selective attention to diagnostic information (e.g. Palmeri & been the case if RTs were directly related to stimulus difficulty.
Gauthier, 2004) can influence how perceptual representations are Second, and crucially, the fact that our findings hold since the first
instantiated (e.g. in IT, Folstein, Gauthier, & Palmeri, 2010). Numer- target presentation, appears to rule out a possible effect of the con-
ous neuroimaging studies have shown that preparation to a partic- struction of the set. It also rules out a possible effect of learning
ular target modulates activity in cortical regions selective to this (repetition of targets and/or a differential effect of adaptation to
target before perception (e.g. Esterman & Yantis, 2010; Gazzaley, different tasks, Fig. 5).
Cooney, McEvoy, Knight, & D’esposito, 2005; Righart, Andersson, Last, the use of a go/no-go paradigm with a response deadline
Schwartz, Mayer, & Vuilleumier, 2009), but also during its presen- prevents to determine whether no-go responses are voluntary or
tation (Esterman & Yantis, 2010; Harel, Kravitz, & Baker, 2014; Kok, missed go-responses. This could have been captured using an
Jehee, & de Lange, 2012). Thus, the activity of these posterior brain old/new paradigm. However, this would have been irrelevant to
areas could be optimized to process target faces. our central aim, which was to assess minimal RT.
The large face inversion effect on Familiar Face Recognition sug-
gests that, even more than Individual Face Recognition, Familiar 4.6. Conclusion
Face Recognition needs face-specific processing. A possible expla-
nation of such a difference is that detecting that a face is familiar Face recognition corresponding to finding a specific person in a
is not possible using individual features only (e.g. an isolated nose crowd appears achievable in only a quarter second. Despite this
is usually not sufficient to recognize someone as familiar), but speed, a hierarchy from Human Face Categorization to Individual
requires an individual-level face representation. Familiar Face Face Recognition and to Familiar Face Recognition was evidenced,
Recognition might therefore rely on longer face-specific processes in favor of the superordinate level, or coarse to fine, hypotheses.
in order to extract the full percept of an individual face (e.g. This study emphasizes that Individual Face Recognition and Famil-
Rossion, 2008; Schiltz & Rossion, 2006). iar Face Recognition are very distinct face recognition tasks and
Familiar Face Recognition and Individual Face Recognition also opens promising directions towards the investigation of how the
rely on very different expectations (i.e. a large number of possible visual system is recruited in each case.
targets on the one hand, one target on the other hand), which may
affect decisional level of processing. Simultaneously, the level of Funding
familiarity may have been different in the two conditions (well-
known targets in the Individual Face Recognition condition, targets This study was supported by Assistance Publique – Hôpitaux de
with variable levels of familiarity, which were not controlled in Marseille – Programme Hospitalier de Recherche Clinique 2001/54
this study, in the Familiar Face Recognition condition). Reducing and France Alzheimer. Gabriel Besson received a grant from the
the variability of the number of targets or increasing the level of foundation Santé, Sport and Développement Durable. These fund-
familiarity could increase the signal-to-noise ratio during Familiar ing sources had no involvement in study design collection, analysis
Face Recognition responses and thus increase the speed of RTs. For and interpretation of data; in the writing of the report; or in the
example, Ramon, Caharel, and Rossion (2011) used a go/no-go task decision to submit the article for publication.
with familiar faces that were all the classmates of the participants
(this task may therefore be lying somewhere between our Familiar
Face Recognition and Individual Face Recognition tasks), who con- Appendix A. Supplementary material
sequently reached high accuracy. However, participants in our
Familiar Face Recognition condition still responded largely faster Supplementary data associated with this article can be found, in
(453 ± 32 ms) than in Ramon et al.’s study (578 ± 18 ms), probably the online version, at http://dx.doi.org/10.1016/j.cognition.2016.
because a response deadline was used. This emphasizes that a pro- 10.004.
cedure such as the SAB can be helpful when assessing the face hier-
archy. It would be interesting to run a task similar to the Ramon’s References
et al.’ study using the SAB.
Anaki, D., & Bentin, S. (2009). Familiarity effects on categorization levels of faces and
Overall, the nature of these two conditions appears largely dif- objects. Cognition, 111(1), 144–149. http://dx.doi.org/10.1016/j.cognition.2009.
ferent regarding the preparation of the visual system, the level of 01.002.
face processing, familiarity and consequently the decisional level Barragan-Jason, G., Besson, G., Ceccaldi, M., & Barbeau, E. J. (2013). Fast and famous:
Looking for the fastest speed at which a face can be recognized. Frontiers in
of processing. This may explain the large difference of RTs observed Psychology, 4. http://dx.doi.org/10.3389/fpsyg.2013.00100.
between them. Barragan-Jason, G., Lachat, F., & Barbeau, E. J. (2012). How fast is famous face
recognition? Frontiers in Psychology, 3, 454. http://dx.doi.org/10.3389/
fpsyg.2012.00454.
4.5. Limits of the study Besson, G., Ceccaldi, M., Didic, M., & Barbeau, E. J. (2012). The speed of visual
recognition memory. Visual Cognition, 20(10), 1131–1152. http://dx.doi.org/
The distinctiveness between targets and distractors was differ- 10.1080/13506285.2012.724034.
Besson, G., Ceccaldi, M., Tramoni, E., Felician, O., Didic, M., & Barbeau, E. J. (2015).
ent between the Human Face Categorization, Individual Face Fast, but not slow, familiarity is preserved in patients with amnestic mild
Recognition and Familiar Face Recognition and may have impacted cognitive impairment. Cortex, 65, 36–49. http://dx.doi.org/10.1016/
reaction times, despite controlling for low-level visual features j.cortex.2014.10.020.
Bowles, B., Crupi, C., Mirsattari, S. M., Pigott, S. E., Parrent, A. G., Pruessner, J. C., ...
(contrast and luminance, pictures cropped and centered around
Köhler, S. (2007). Impaired familiarity with preserved recollection after anterior
faces, distractors paired one-by-one with targets for the Individual temporal-lobe resection that spares the hippocampus. In Proceedings of the
Face Recognition conditions). These differences between condi- national academy of sciences (Vol. 104, pp. 16382–16387).
tions are inherent to the Human Face Categorization, Individual Brown, V., Huey, D., & Findlay, J. M. (1997). Face detection in peripheral vision: do
faces pop out? Perception, 26(12), 1555–1570.
Face Recognition and Familiar Face Recognition conditions and Bruce, V., & Young, A. (1986). Understanding face recognition. British Journal of
hardly reducible. Nonetheless, different results suggest that our Psychology, 77(3), 305–327.
 G. Besson et al. / Cognition 158 (2017) 33–43 43

Carey, S., & Diamond, R. (1977). From piecemeal to configurational representation of Palmeri, T. J., & Gauthier, I. (2004). Visual object understanding. Nature Reviews.
faces. Science (New York, NY), 195(4275), 312–314. Neuroscience, 5(4), 291–303. http://dx.doi.org/10.1038/nrn1364.
Carey, S., Schonen, S. D., & Ellis, H. D. (1992). Becoming a face expert [and Poncet, M., & Fabre-Thorpe, M. (2014). Stimulus duration and diversity do not
discussion]. Philosophical Transactions of the Royal Society B: Biological Sciences, reverse the advantage for superordinate-level representations: The animal is
335(1273), 95–103. http://dx.doi.org/10.1098/rstb.1992.0012. seen before the bird. The European Journal of Neuroscience, 39(9), 1508–1516.
Cauchoix, M., Barragan-Jason, G., Serre, T., & Barbeau, E. J. (2014). The neural http://dx.doi.org/10.1111/ejn.12513.
dynamics of face detection in the wild revealed by MVPA. The Journal of Praß, M., Grimsen, C., König, M., & Fahle, M. (2013). Ultra rapid object
Neuroscience: The Official Journal of the Society for Neuroscience, 34(3), 846–854. categorization: effects of level. Animacy and Context. PLoS ONE, 8(6), e68051.
http://dx.doi.org/10.1523/JNEUROSCI.3030-13.2014. http://dx.doi.org/10.1371/journal.pone.0068051.
Crouzet, S. M., Kirchner, H., & Thorpe, S. (2010). Fast saccades toward faces: Face Ramon, M., Caharel, S., & Rossion, B. (2011). The speed of recognition of personally
detection in just 100 ms. Journal of Vision, 10(4), 1–17. http://dx.doi.org/ familiar faces. Perception, 40(4), 437–449. http://dx.doi.org/10.1068/p6794.
10.1167/10.4.16. Reddy, L., Reddy, L., & Koch, C. (2006). Face identification in the near-absence of
Delorme, A., & Thorpe, S. J. (2001). Face identification using one spike per neuron: focal attention. Vision Research, 46(15), 2336–2343. http://dx.doi.org/10.1016/j.
resistance to image degradations. Neural Networks, 14(6), 795–803. visres.2006.01.020.
Dosher, B. A. (1976). The retrieval of sentences from memory: A speed-accuracy Reed, A. V. (1973). Speed-Accuracy Trade-Off in Recognition Memory. Science, 181,
study. Cognitive Psychology, 8, 291–310. 574–576.
Eimer, M. (2014). The neural basis of attentional control in visual search. Trends in Righart, R., Andersson, F., Schwartz, S., Mayer, E., & Vuilleumier, P. (2009). Top-
Cognitive Sciences, 18(10), 526–535. http://dx.doi.org/10.1016/j.tics.2014. down activation of fusiform cortex without seeing faces in prosopagnosia.
05.005. Cerebral Cortex, 20(8), 1878–1890. http://dx.doi.org/10.1093/cercor/bhp254.
Esterman, M., & Yantis, S. (2010). Perceptual expectation evokes category-selective Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic
cortical activity. Cerebral Cortex, 20(5), 1245–1253. http://dx.doi.org/10.1093/ objects in natural categories. Cognitive Psychology, 8(3), 382–439.
cercor/bhp188. Rossion, B. (2008). Picture-plane inversion leads to qualitative changes of face
Fabre-Thorpe, M. (2011). The characteristics and limits of rapid visual perception. Acta Psychologica, 128(2), 274–289. http://dx.doi.org/10.1016/j.
categorization. Frontiers in Psychology, 2. http://dx.doi.org/10.3389/fpsyg. actpsy.2008.02.003.
2011.00243. Rossion, B., & Caharel, S. (2011). ERP evidence for the speed of face categorization in
Farah, M. J., Wilson, K. D., Drain, M., & Tanaka, J. N. (1998). What is ‘‘ special” about the human brain: Disentangling the contribution of low-level visual cues from
face perception? Psychological Review, 105(3), 482. face perception. Vision Research, 51(12), 1297–1311. http://dx.doi.org/10.1016/j.
Folstein, J. R., Gauthier, I., & Palmeri, T. J. (2010). Mere exposure alters category visres.2011.04.003.
learning of novel objects. Frontiers in Psychology, 1, 40. http://dx.doi.org/ Rossion, B., & Jacques, C. (2011). The N170: Understanding the time course of face
10.3389/fpsyg.2010.00040. perception in the human brain. In The oxford handbook of event-related potential
Gazzaley, A., Cooney, J., McEvoy, K., Knight, R., & D’esposito, M. (2005). Top-down components (pp. 115).
enhancement and suppression of the magnitude and speed of neural activity. Rousselet, G. A., Macé, M. J.-M., & Fabre-Thorpe, M. (2003). Is it an animal? Is it a
Journal of Cognitive Neuroscience, 17(3), 507–517. human face? Fast processing in upright and inverted natural scenes. Journal of
Goffaux, V., Peters, J., Haubrechts, J., Schiltz, C., Jansma, B., & Goebel, R. (2011). From Vision, 3(6).
coarse to fine? Spatial and temporal dynamics of cortical face processing. Russell, R., Duchaine, B., & Nakayama, K. (2009). Super-recognizers: People with
Cerebral Cortex, 21(2), 467–476. http://dx.doi.org/10.1093/cercor/bhq112. extraordinary face recognition ability. Psychonomic Bulletin & Review, 16(2),
Harel, A., Kravitz, D. J., & Baker, C. I. (2014). Task context impacts visual object 252–257. http://dx.doi.org/10.3758/PBR.16.2.252.
processing differentially across the cortex. Proceedings of the National Academy Schiltz, C., & Rossion, B. (2006). Faces are represented holistically in the human
of Sciences, 111(10), E962–E971. http://dx.doi.org/10.1073/pnas.1312567111. occipito-temporal cortex. NeuroImage, 32(3), 1385–1394. http://dx.doi.org/
Hochstein, S., & Ahissar, M. (2002). View from the top: Hierarchies and reverse 10.1016/j.neuroimage.2006.05.037.
hierarchies in the visual system. Neuron, 36(5), 791–804. Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory:
Jacques, C., & Rossion, B. (2006). The speed of individual face categorization. Applications to dementia and amnesia. Journal of Experimental Psychology:
Psychological Science, 17(6), 485–492. http://dx.doi.org/10.1111/j.1467-9280. General, 117(1), 34.
2006.01733.x. Sugase, Y., Yamane, S., Ueno, S., & Kawano, K. (1999). Global and fine information
Johnson, K. E., & Mervis, C. B. (1997). Effects of varying levels of expertise on the coded by single neurons in the temporal visual cortex. Nature, 400(6747),
basic level of categorization. Journal of Experimental Psychology. General, 126(3), 869–873. http://dx.doi.org/10.1038/23703.
248–277. Tanaka, J. W. (2001). The entry point of face recognition: Evidence for face expertise.
Jolicoeur, P., Gluck, M. A., & Kosslyn, S. M. (1984). Pictures and names: Making the Journal of Experimental Psychology. General, 130(3), 534–543.
connection. Cognitive Psychology, 16(2), 243–275. Tanaka, J., & Gauthier, I. (1997). Expertise in object and face recognition. Psychology
Kadar, I., & Ben-Shahar, O. (2012). A perceptual paradigm and psychophysical of Learning and Motivation (Vol. 36, pp. 83–125). Elsevier.
evidence for hierarchy in scene gist processing. Journal of Vision, 12(13), 16. Tanaka, J. W., & Taylor, M. (1991). Object categories and expertise: Is the basic level
http://dx.doi.org/10.1167/12.13.16. in the eye of the beholder? Cognitive Psychology, 23(3), 457–482. http://dx.doi.
Kalaska, J. F., & Crammond, D. J. (1992). Cerebral cortical mechanisms of reaching org/10.1016/0010-0285(91)90016-H.
movements. Science (New York, NY), 255(5051), 1517–1523. Tsao, D. Y., & Livingstone, M. S. (2008). Mechanisms of face perception. Annual
Kok, P., Jehee, J. F. M., & de Lange, F. P. (2012). Less is more: Expectation sharpens Review of Neuroscience, 31(1), 411–437. http://dx.doi.org/10.1146/annurev.
representations in the primary visual cortex. Neuron, 75(2), 265–270. http://dx. neuro.30.051606.094238.
doi.org/10.1016/j.neuron.2012.04.034. Valentine, T. (1988). Upside-down faces: A review of the effect of inversion upon
Large, M.-E., Kiss, I., & McMullen, P. A. (2004). Electrophysiological correlates of face recognition. British Journal of Psychology (London, England: 1953), 79(Pt 4),
object categorization: Back to basics. Cognitive Brain Research, 20(3), 415–426. 471–491.
http://dx.doi.org/10.1016/j.cogbrainres.2004.03.013. Valentine, T., & Bruce, V. (1988). Mental rotation of faces. Memory & Cognition, 16(6),
Lewis, M. B., & Ellis, H. D. (2000). Satiation in name and face recognition. Memory & 556–566.
Cognition, 28, 783–788. Valentine, T. (2001). Face-space models of face recognition. In M. J. Wenger & J. T.
Loschky, L. C., & Larson, A. M. (2010). The natural/man-made distinction is made Townsend (Eds.), Computational, geometric, and process perspectives on facial
before basic-level distinctions in scene gist processing. Visual Cognition, 18(4), cognition: Contexts and challenges. Hillsdale, New Jersey: Lawrence Erlbaum
513–536. http://dx.doi.org/10.1080/13506280902937606. Associates Inc..
Löw, A., Bentin, S., Rockstroh, B., Silberman, Y., Gomolla, A., Cohen, R., & Elbert, T. Vanmarcke, S., Van Der Hallen, R., Evers, K., Noens, I., Steyaert, J., & Wagemans, J.
(2003). Semantic categorization in the human brain spatiotemporal dynamics (2016). Ultra-rapid categorization of meaningful real-life scenes in adults with
revealed by magnetoencephalography. Psychological Science, 14(4), 367–372. and without ASD. Journal of Autism and Developmental Disorders, 46, 450–466.
Macé, M. J.-M., Joubert, O. R., Nespoulous, J.-L., & Fabre-Thorpe, M. (2009). The time- Vanmarcke, S., & Wagemans, J. (2015). The influence of age and gender on ultra-
course of visual categorizations: You spot the animal faster than the bird. PLoS rapid categorization. Visual Cognition, 23, 894–916.
ONE, 4(6), e5927. http://dx.doi.org/10.1371/journal.pone.0005927. VanRullen, R., Reddy, L., & Koch, C. (2004). Visual search and dual tasks reveal two
Mack, M. L., Gauthier, I., Sadr, J., & Palmeri, T. J. (2008). Object detection and basic- distinct attentional resources. Journal of Cognitive Neuroscience, 16(1), 4–14.
level categorization: Sometimes you know it is there before you know what it VanRullen, R., & Thorpe, S. J. (2001). Is it a bird? Is it a plane? Ultra-rapid visual
is. Psychonomic Bulletin & Review, 15, 28–35. categorisation of natural and artifactual objects. Perception, 30(6), 655–668.
Mack, M. L., & Palmeri, T. J. (2011). The timing of visual object categorization. http://dx.doi.org/10.1068/p3029.
Frontiers in Psychology, 2. http://dx.doi.org/10.3389/fpsyg.2011.00165. Wong, A. C.-N., & Gauthier, I. (2007). An analysis of letter expertise in a levels-of-
Martinovic, J., Gruber, T., & Müller, M. M. (2008). Coding of visual object features categorization framework. Visual Cognition, 15(7), 854–879. http://dx.doi.org/
and feature conjunctions in the human brain. PLoS ONE, 3(11), e3781. http://dx. 10.1080/13506280600948350.
doi.org/10.1371/journal.pone.0003781. Yin, R. K. (1969). Looking at upside-down faces. Journal of Experimental Psychology,
Maurer, D., Grand, R. L., & Mondloch, C. J. (2002). The many faces of configural 81(1), 141–145. http://dx.doi.org/10.1037/h0027474.
processing. Trends in Cognitive Sciences, 6(6), 255–260.