Journal of Vision 2007 7 10 6 1 13 http journalofvision org 7 10 6 1 A bottom up model of spatial attention predicts human error patterns in rapid scene recognition Wolfgang Einh user Division of Biology California Institute of Technology Pasadena CA USA T Nathan Mundhenk Department of Computer Science University of Southern California Los Angeles CA USA Pierre Baldi Christof Koch School of Information and Computer Sciences University of California in Irvine Irvine CA USA Division of Biology California Institute of Technology Pasadena CA USA Department of Computer Science University of Southern California Los Angeles CA USA Laurent Itti Humans demonstrate a peculiar ability to detect complex targets in rapidly presented natural scenes Recent studies suggest that nearly no focal attention is required for overall performance in such tasks Little is known however of how detection performance varies from trial to trial and which stages in the processing hierarchy limit performance bottom up visual processing attentional selection and or recognition or top down factors e g decision making memory or alertness uctuations To investigate the relative contribution of these factors eight human observers performed an animal detection task in natural scenes presented at 20 Hz Trial by trial performance was highly consistent across observers far exceeding the prediction of independent errors This consistency demonstrates that performance is not primarily limited by idiosyncratic factors but by visual processing Two statistical stimulus properties contrast variation in the target image and the information theoretical measure of surprise in adjacent images predict performance on a trial by trial basis These measures are tightly related to spatial attention demonstrating that spatial attention and rapid target detection share common mechanisms To isolate the causal contribution of the surprise measure eight additional observers performed the animal detection task in sequences that were reordered versions of those all subjects had correctly recognized in the rst experiment Reordering increased surprise before and or after the target while keeping the target and distractors themselves unchanged Surprise enhancement impaired target detection in all observers Consequently and contrary to several previously published ndings our results demonstrate that attentional limitations rather than target recognition alone affect the detection of targets in rapidly presented visual sequences Keywords psychophysics modeling attention saliency RSVP Citation Einh user W Mundhenk T N Baldi P Koch C Itti L 2007 A bottom up model of spatial attention predicts human error patterns in rapid scene recognition Journal of Vision 7 10 6 1 13 http journalofvision org 7 10 6 doi 10 1167 7 10 6 Introduction Humans and other primates grasp the gist of a complex natural scene even when presented for only a few tens of milliseconds Biederman 1981 Evans Treisman 2005 Fabre Thorpe Richard Thorpe 1998 Li VanRullen Koch Perona 2002 Potter Levy 1969 Rousselet Fabre Thorpe Thorpe 2002 Thorpe Fize Marlot 1996 VanRullen Thorpe 2001 Furthermore observers can detect with above chance doi 1 0 11 67 7 1 0 6 performance complex target items such as an animal in rapidly presented image sequences rapid serial visual presentation RSVP Evans Treisman 2005 Potter Levy 1969 Such performance is typically seen as evidence for a rapid sensory driven bottom up mode of processing primarily driven by the visual stimulus This leads to the hypothesis that properties of the stimulus rather than observer specific and possibly more idiosyncratic top down processes may to a large extent determine performance in RSVP If so what are these statistical properties Received December 3 2006 published June 20 2007 ISSN 1534 7362 ARVO Journal of Vision 2007 7 10 6 1 13 Einh user et al It has been argued that rapid recognition requires little or no focal spatial attention Li et al 2002 Rousselet et al 2002 According to this view bottom up attention does not constitute the primary limit for rapid visual processing but rather such a limit is found in a later target recognition stage Indeed some aspects of overall performance can be captured by models of object recognition for example animals that appear farther away are more difficult to detect on average Serre Oliva Poggio 2006 However these studies typically use isolated images followed by masking stimuli Contrary to these results when using a stream of images some of which are targets and most of which act as distractors one finds two attentional phenomena that limit rapid processing When two identical items are presented in direct succession often only one is detected repetition blindness Kanswisher 1987 and when a second target item is presented shortlyVbut not immediatelyVafter a first one its processing is also impaired attentional blink Raymond Shapiro Arnell 1992 Although repetition blindness and attentional blink are distinct phenomena Chun 1997 models of such attentional impairments are typically variants of an attentional gating model as first formalized by Reeves and Sperling 1986 In this view a salient item e g a target opens an attentional gate for its and subsequent items access to visual short term memory Failure to quickly reopen the gate impairs the detection of the second target in attentional blink furthermore integration of information according to order and strength within an open gate epoch leads to the loss of order information a potential cause for repetition blindness In attentional blink the saliency of an item to open a gate arises from its property of being a target or semantically related to the target Barnard Scott Taylor May Knightley 2004 Items that attract attention because of their emotional content can also lead to an attentional blink like recognition impairment which some but not all observers can overcome through volitional control Most Chun Widders Zald 2005 Similarly odd items e g the rare occurrence of a face in a letter task or vice versa can impair subsequent processing Marois Todd Gilbert 2003 as can items that are visually similar to the target but appear at peripheral locations Folk Leber Egeth 2002 However very little is known quantitatively of the neural mechanisms by which some items may strongly capture attention and create an attentional blink like effect We hypothesize that high stimulus driven saliency can impair the detection of subsequent targets In the view of
View Full Document
Unlocking...