Purchase this article with an account.
Alexander Bratch, Aparna Srinath, Shawn Barr, William Bromfield, Jason Gold; Configural superiority reduces efficiency. Journal of Vision 2014;14(10):244. doi: https://doi.org/10.1167/14.10.244.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The ability of human observers to recognize visual patterns can be greatly influenced by context. For example, the introduction of context to a set of features can induce a unified percept, allowing observers to classify it more quickly and accurately (a configural superiority effect; Pomerantz & Portillo, 2011). But how does context specifically impact the way in which observers make use of stimulus information? We addressed this question by applying ideal observer analysis (Geisler, 1989) and response classification (Ahumada, 2002) to a pattern identification task that has been shown previously to exhibit a significant configural superiority effect (Pomerantz & Portillo, 2011). In the No Context condition, four diagonal lines appeared in separate quadrants of a virtual square. One of these lines was placed at a countering angle, and the observers task was to identify which randomly chosen quadrant contained the odd angle. In the Context condition, the stimuli were identical, with the exception that two abutting line segments (one vertical and one horizontal, forming a right angle) were added in the same position to all four quadrants. Thus, the additional features were entirely redundant, but their presence induced the percept of a unitary triangle figure within the quadrant containing the odd angle. On each trial, the stimulus was embedded in Gaussian white contrast noise, which allowed us to measure human and ideal observer contrast energy identification thresholds, as well as the pixel-wise information-processing strategy adopted by observers in each condition. Although response times were faster in the presence of context, efficiency (ideal/human threshold) was actually lower when the stimulus included the redundant features. The results of the response classification analyses suggest that the lower efficiencies in the presence of context were in part due to observers reliance on the redundant features, which contribute noise while providing no discriminative information.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only