Eye-Tracking as a Lens into Expertise Development in Visual Search

Abstract

Visual search professionals frequently examine complex images in which they must locate and identify anomalies (e.g., tumors or lesions in medical images, hostile territories in maps) indicating the presence of important information. This is a difficult perceptual and cognitive challenge that requires years of experience even beyond domain-specific training. We developed a laboratory analogue of this task to examine how people development expertise over time and the changes that occur in visual scanning behaviors as this expertise accrues. Participants' eyes were tracked as they searched for subtle anomalies during and after different forms of training. Anomaly detection improved with experience and this improvement occurred more quickly in perceptual training conditions. Eye movement analyses revealed that participants' expertise conferred benefits in both scanning and recognition times. These results suggest that simple perceptual training methods can affect both cognitive and oculomotor components of visual search.

Keywords: ACM proceedings; Clutter; Expertise; Image Perception; Visual Search.

Figure 1:

Examples of a typical laboratory visual search array (left; from [21]), an abdominal CT scan (middle; from James Heilman, MD, Wikimedia Commons), and an aerial map (https://pxhere.com/en/photo/919491).

Figure 2:

Example scenes from the ODDS Database [15]. Rows indicate different scenes and columns indicate different subtlety ratings for anomalies embedded in those scenes. Embedded yellow arrows indicate the location of anomalous elements.

Figure 3:

Timeline of the present investigation.

Figure 4:

Schematic of each session as participants developed task expertise.

Figure 5:

Effect of subtlety rating on the probability of recognition errors in the array search task (blue shading reflects standard error)

Figure 6:

Effect of subtlety rating on the probability of recognition errors in the scene search task (blue shading reflects standard error)

Figure 7:

Effect of subtlety rating on dwell time in the array search task (error bars depict standard error)

Figure 8:

Effect of subtlety rating on dwell time in the scene search task (error bars depict standard error)