Attention, Perception, & Psychophysics
2009, 71 (3), 594-606
doi:10.3758/APP.71.3.594
Processing of natural images is feedforward:
A simple behavioral test
Thomas schmidT and Filipp schmidT
University of Giessen, Giessen, Germany
Natural images can be classified so rapidly that it has been suggested that their analysis is based on a first
single pass of processing activity through the visuomotor system. We tested this theory in a visuomotor priming
task in which speeded pointing responses were performed toward one of two target images containing a prespeci-
fied stimulus (e.g., animal vs. nonanimal, ellipse vs. rectangle). Target pictures were preceded by prime pictures
of the same or an opposite category, linked to either the same or an opposite pointing response. We found that
pointing trajectories were initially controlled by the primes alone, but independently of information in the actual
targets. Our data indicate that prime and target signals remained strictly sequential throughout all processing
stages, meeting unprecedentedly stringent behavioral criteria for feedforward processing (rapid-chase criteria).
Our findings suggest that visuomotor priming effects capture the output of the very first pass of information
through the visuomotor system, before output is affected by recurrent information.
Our visual system is able to categorize images of natural visuomotor system (VanRullen & Thorpe, 2001, 2002).
visual scenes at a remarkable speed (Hegdé, 2008). Evi- In other words, categorization of natural images, as well
dence for this has come from studies in which participants as the translation into appropriate motor responses, may
were presented with single images and had to indicate as be completed within a fast feedforward sweep of visuo-
quickly as possible whether or not the image contained motor processing (Bullier, 2001; Lamme & Roelfsema,
an animal, using a go/no-go response (e.g., Thorpe, Fize, 2000). In the context of neuronal signal flow, feedforward
& Marlot, 1996; VanRullen & Koch, 2003; VanRullen & indicates that a cell passes activation on to another cell
Thorpe, 2001, 2002). More recent studies have employed before integrating any feedback or recurrent information
a two-alternative forced choice paradigm in which two from other cells about that signal. During the fast feed-
pictures were presented side by side and participants had forward sweep, a wavefront of visually elicited activation
to indicate which one contained the animal (see Bacon- would travel through the visuomotor system so fast that it
Macé, Kirchner, Fabre-Thorpe, & Thorpe, 2007, for a would essentially be devoid of information from recurrent
comparison of both types of task). The most spectacular processing, which would develop only in the wake of the
results in this variant of the task have come from a study wave (Lamme & Roelfsema, 2000). Simulation studies
that required participants to make a rapid eye movement of rapid stimulus classification in artificial neuronal net-
toward the animal picture (Kirchner & Thorpe, 2006). works indeed have suggested that most of the stimulus-
These authors found that the rate of correct responses relevant information could be extracted from the tempo-
began to exceed the rate of errors at saccade latencies as ral distribution of the very first spikes in the feedforward
short as 120 msec. Importantly, all of the studies men- wavefront (Serre, Oliva, & Poggio, 2007; VanRullen, De-
tioned above were able to trace the transition from nondis- lorme, & Thorpe, 2001; VanRullen, Gautrais, Delorme, &
criminative to discriminative processing in the response Thorpe, 1998). The issue of feedforward versus recurrent
time distributions of the system’s motor output: The fast- processing is theoretically interesting because many au-
est responses in distributions are still as likely to be cor- thors have assumed that feedforward processing alone is
rect as to be incorrect (e.g., Kirchner & Thorpe, 2006; insufficient to generate visual awareness and that a stimu-
VanRullen & Koch, 2003), and the earliest segments of lus must be processed recurrently to become consciously
the response time function are unaffected by subsequent accessible (DiLollo, Enns, & Rensink, 2000; Fahren-
visual masking of the target images (Bacon-Macé, Macé, fort, Scholte, & Lamme, 2007; Lamme, 2002; Lamme,
Fabre-Thorpe, & Thorpe, 2005). These findings suggest Rodriguez-Rodriguez, & Spekreijse, 1999; Lamme &
that response activation is indeed very fast and direct. Roelfsema, 2000; Lamme, Zipser, & Spekreijse, 2002;
Because of the sheer rapidity of image classification, Pascual-Leone & Walsh, 2001; Ro, Breitmeyer, Burton,
it has been argued that natural scene processing has to Singhal, & Lane, 2003; Roelfsema, Tolboom, & Khayat,
occur during the first pass of informati