Visual Attention and Change Detection Ty W. Boyer (tywboyer@indiana.edu) Thomas G. Smith (thgsmith@indiana.edu) Chen Yu (chenyu@indiana.edu) Bennett I. Bertenthal (bbertent@indiana.edu) Department of Psychological & Brain Sciences, Indiana University 1101 East Tenth Street Bloomington, IN 47405 USA Abstract Studies suggest that visual attention, guided in part by features’ visual salience, is necessary for change detection. An image processing algorithm was used for measuring the visual salience of the features of scenes, and participants’ ability to detect changes made to high and low salience features was measured with a flicker paradigm while their eye movements were recorded. Changes to high salience features were fixated sooner, for shorter durations, and were detected faster and with higher accuracy than those made to low salience features. The implications of these results for visual attention and change detection research are discussed. Keywords: change detection; visual salience; eye tracking Introduction Attention is the process of allocating perceptual and cognitive resources to select which information in the environment will enter consciousness. Finding a person in a crowd, locating one’s car in a parking garage, identifying features that distinguish predator from prey, or finding a red circle amongst red and blue squares in a laboratory experiment all require control of visual attention (Kahneman & Henik, 1981; Triesman & Gelade, 1980; Wolfe, 2003). Attention contributes to an ability to make sense of our rich visual world and learn from experience. The degree to which attention is guided by the features of the viewed stimulus versus the viewer’s goals, expectations, and subjective evaluations are of paramount importance to researchers who study visual attention (Egeth & Yantis, 1997; Torralba, Oliva, Castelhano, & Henderson, 2006; Treue, 2003). Recently, computational saliency models have been developed to analyze visual scenes in terms of their stimulus properties (Itti, Koch, & Niebur, 1998; Koch & Ullman, 1985; Parkhurst, Law, & Niebur, 2002). These models have been used to predict viewers’ fixation patterns as they view images, providing support for the suggestion that bottom-up visual saliency contributes to the guidance of visual attention (Itti & Koch, 2001; Foulsham & Underwood, 2008). Other research shows that in the absence of visual attention we are particularly poor at detecting changes made to the features of scenes, a phenomenon known as change blindness (Hollingworth, 2006; Rensink, 2000a; 2000b; Rensink, O’Regan, & Clark, 1997; Simons & Levin, 1997). Changes that occur during saccades, eye-blinks, interleaved frames, cuts, and pans largely escape perceptual awareness. In experimental change detection tasks, visual salience is one factor guiding the direction of attention to features in the scene, and thus it is conjectured that salience contributes to whether and how quickly the changing feature will be detected (Kelley, Chun, & Chua, 2003; Simons & Ambinder, 2005). This is suggested under the assumption that viewers must direct visual attention to the feature that is changing, and are unlikely to do so if it is less salient than other features competing for visual attention. Yet, contrary to this prediction, two recent studies, Stirk and Underwood (2007) and Wright (2005), report that the visual salience of stimulus image features, determined with formal salience algorithms, does not predict response times in a change detection task. By contrast, both of these studies found that the higher level semantic characteristics of changing features influenced their detection speeds (i.e., the changing feature’s congruence with the theme of the scene, or whether it had been subjectively rated as high or low salience by independent viewers). Neither of these studies directly measured visual attention (i.e., eye movements), however, so it is uncertain how salience may have affected visual attention. As such, why these previous studies failed to find a relation between salience and change detection in requires closer scrutiny. In the current research, eye movements are used to systematically assess the distribution of attention across each change detection trial. The primary goals of the current study are to examine: 1) whether stimulus feature salience predicts visual attention in a change detection paradigm, and 2) whether change detection requires overt visual attention, or whether covert attention suffices. Computational saliency maps were used to identify the visual salience of the features within a set of images (Itti et al., 1998). Changes were applied to features identified as either high or low salience, and participants viewed these modified images interleaved with the originals in a flicker paradigm (Rensink et al., 1997). Visual attention was measured with a remote eye-tracking system that enabled examination of the fixation sequences that index overt visual attention, as well as the fixation durations that index the amount of cognitive processing of features during the search process. In Experiment 1 participants viewed scenes until they executed a manual response indicating change detection. Since the amount of time they had to view the scenes was open-ended, this experiment is limited in its ability to determine whether overt attention is necessary or
[1]
Antonio Torralba,et al.
Contextual guidance of eye movements and attention in real-world scenes: the role of global features in object search.
,
2006,
Psychological review.
[2]
Ronald A. Rensink.
The Dynamic Representation of Scenes
,
2000
.
[3]
C. Koch,et al.
Computational modelling of visual attention
,
2001,
Nature Reviews Neuroscience.
[4]
A. Hollingworth.
Visual memory for natural scenes: Evidence from change detection and visual search
,
2006
.
[5]
S. Treue.
Visual attention: the where, what, how and why of saliency
,
2003,
Current Opinion in Neurobiology.
[6]
Derrick J. Parkhurst,et al.
Modeling the role of salience in the allocation of overt visual attention
,
2002,
Vision Research.
[7]
Gregory J. Zelinsky,et al.
Scene context guides eye movements during visual search
,
2006,
Vision Research.
[8]
A. Torralba,et al.
The role of context in object recognition
,
2007,
Trends in Cognitive Sciences.
[9]
Christof Koch,et al.
A Model of Saliency-Based Visual Attention for Rapid Scene Analysis
,
2009
.
[10]
M. Wright.
Saliency predicts change detection in pictures of natural scenes.
,
2005,
Spatial vision.
[11]
S Ullman,et al.
Shifts in selective visual attention: towards the underlying neural circuitry.
,
1985,
Human neurobiology.
[12]
Ronald A. Rensink.
Seeing, sensing, and scrutinizing
,
2000,
Vision Research.
[13]
A. Treisman,et al.
A feature-integration theory of attention
,
1980,
Cognitive Psychology.
[14]
I. Rock,et al.
Perceptual organization and attention
,
1992,
Cognitive Psychology.
[15]
G. Underwood,et al.
Low-level visual saliency does not predict change detection in natural scenes.
,
2007,
Journal of vision.
[16]
Ronald A. Rensink,et al.
TO SEE OR NOT TO SEE: The Need for Attention to Perceive Changes in Scenes
,
1997
.
[17]
T. Foulsham,et al.
What can saliency models predict about eye movements? Spatial and sequential aspects of fixations during encoding and recognition.
,
2008,
Journal of vision.
[18]
Michael S. Ambinder,et al.
Change blindness
,
1997,
Trends in Cognitive Sciences.
[19]
S. Yantis,et al.
Visual attention: control, representation, and time course.
,
1997,
Annual review of psychology.
[20]
T. A. Kelley,et al.
Effects of scene inversion on change detection of targets matched for visual salience.
,
2003,
Journal of vision.