Abstract
Previous work with high- and low-spatial frequency hybrid images has demonstrated a bias for extracting information at the spatial frequency scale diagnostic of the viewing task. Specifically, the visual system can be influenced by top-down information to flexibly select the appropriate task-dependent (diagnostic) spatial frequency channel. These results have been interpreted to suggest cognitive penetrability of early perceptual processes (Oliva & Schyns, 1997; Schyns & Oliva, 1999). The current study sought to determine whether gaze selection is similarly flexible. High- and low-spatial frequency hybrid images were constructed from two superimposed faces (one at each scale) such that regions of interest (eyes, nose, mouth) did not overlap. In the first experiment, subjects were instructed to identify (by name) either the high or low spatial frequency (HSF or LSF) face in the hybrid. Results suggest that the gaze control system is most effective at targeting HSF information for selection regardless of which spatial frequency is most task relevant. Based on these results, we reject the strong hypothesis that the gaze control system is maximally flexible in selecting only information in the task-dependent spatial frequency channel. However, it is possible that these results may reflect either a specific tuning to the scale most likely to contain diagnostic information (i.e., HSF information for face recognition), or alternatively, a general bias to select HSF information. In the second experiment, we tested these alternative explanations. Participants were asked to categorize the mood expression of each face at one of two spatial frequency scales - a task that has been previously shown to rely heavily on cues from LSF scales. We discuss the implications of our results in relation to the cognitive penetrability of gaze selection.
This work was supported by the National Science Foundation (BCS-0094433 and ECS-9873531, NSF IGERT Program (DGE0114378), and the Army Research Office (W911NF-04-1-0078).