Abstract
Introduction: Numerous studies have shown that humans make the vast majority of their fixations around the eye region when looking at a face (e.g., Barton et al., 2006). We have previously shown that this region contains a rich amount of information for identification (Peterson & Eckstein, 2008). However, humans accomplish many tasks with faces. Here, we investigated performance as a function of fixation location across four common face tasks and asked whether a simple eye movement strategy approximates optimality. Methods: Stimuli consisted of grayscale faces embedded in white Gaussian noise. The Identification condition contained ten neutral faces. The Gender condition contained 80 faces (40 female). The Emotion condition contained 140 faces (20 each of the 7 basic emotions; 70 female). The Happy-Neutral condition contained 100 faces (50 happy, 50 neutral). In each of 1500 trials observers fixated one of five randomly sampled points along the vertical midline of the face. A face was shown for 200 ms followed by a response screen. We derived an ideal observer fitted with a human-like foveated visual system. Model parameters were fit to human data from the Identification condition and used to simulate predicted performance for the remaining tasks. Results: Human performance showed a remarkably consistent pattern across tasks, with performance peaking for fixations around the eyes and remaining relatively unchanged towards the nose tip. Foveated ideal observer results also predicted a peak toward the eyes with a relatively unchanged performance profile toward the nose followed by a steep drop-off above the eyes and below the nose tip. Conclusion: Natural systems analysis (Geisler, 2009) reveals that the combination of the foveated nature of the human visual system and the structure of the human face leads to a simple strategy of fixating the eye region being a rational approach toward maximizing performance on many common face-related tasks.