查看文章

The majority of work in human and machine vision to date has made the simplifying assumption that visual acuity during stimulus input is equally good across the image to be processed. A property of human perception, though, is that high acuity vision is restricted to a small (2) foveal region surrounding fixation, with acuity dropping off precipitously from the fixation point (Anstis. 1974; Riggs. 1965). The human visual system takes advantage of this high-acuity region by rapidly reorienting the eyes via very fast (saccadic) eye movements (Buswell. 1935; Henderson & Hollingworth. 1998, 1999; Rayner, 1998; Yarbus. 1967). Recent work in computer vision and robotics (Kuniyoshi et al.. 1995; Brooks et al.. 1998) suggests that outfitting artificial vision systems with a central high-acuity region can similarly provide important computational advantages in computer vision. However, foveated vision systems require that the direction of gaze be controlled so that the foveal region is appropriately directed within the image based on the properties of the stimulus and the goals of the agent (human or machine), a complex real-time learning and control problem. The interdisciplinary project described in this chapter is an attempt to integrate the study of human and machine gaze control, with the ultimate goal of shedding light on the underlying principles and properties of gaze control within the important context of face perception.

Gaze Control in Human Vision. The human visual system takes advantage of the high resolving power of the fovea by reorienting the fixation point around the viewed scene an average of three times each second via saccadic eye …