Measuring the flexibility of orientation selectivity in face processing by varying task demands.

Observers preferentially process information conveyed by the horizontal orientation band when identifying faces (Dakin and Watt, J Vis 2009; Goffaux and Dakin, Front Psychol 2010). However, ideal observer analysis reveals that such horizontal selectivity is optimal in face identification tasks (Pachai et al, Front Psychol 2013). Therefore, it remains unclear whether horizontal selectivity results from a flexible system tuned to the most diagnostic band for a given task, or a general bias present during all face-related tasks. To disambiguate these hypotheses, we asked observers to perform two face-related tasks for which the diagnostic orientation band differed. On each trial, one of six identities was presented with the head turned slightly to the left or right. Observers were asked on different trials, either blocked or intermixed, to judge the stimulus identity or viewpoint direction. Stimuli were masked with high-contrast orientation-filtered noise (horizontal or vertical, bandwidth = 90 deg) and a low-contrast white noise to enable ideal observer analysis. The dependent measure was the d'=1 RMS contrast threshold, which should be elevated from baseline proportionally to the weight placed by the observer on the masked orientation band during the task in question. A simulated ideal observer confirmed the differential diagnosticity of orientation bands in the two tasks: more masking produced by horizontal noise in the face identification task, and more masking produced by vertical noise in the viewpoint direction task. However, human observers exhibited more masking from horizontal noise in both the identity and direction tasks, regardless of whether these tasks were blocked or intermixed. This result demonstrates an inability to preferentially process vertical facial structure even when it is optimal for the task at hand, and suggests that horizontal selectivity may represent a general face processing strategy. Meeting abstract presented at VSS 2015.