Abstract
A study on gaze perception was described to exemplify image processing and presentation techniques that are useful for psychophysical experiments. We perceive another person's gaze direction accurately from straight ahead, but the perception becomes inaccurate when the person's head is turned. This error has been attributed to the misjudgment of the gaze itself, but the error could stem from another source. To examining such a possibility we conducted several experiments. First, we measured the accuracy of gaze-direction estimation for full-size color photographs with several gaze and head-turn angles presented on CRT. Results indicate that gaze perception is accurate from straight ahead, but that a small but systematic error occurs for stimuli with a head turn. Next, we had subjects evaluate head turns for similar stimuli, and found that the head turn is generally underestimated. The results suggest that the error in estimating gaze direction is caused by underestimation of head turn. This possibility was further tested by presenting synthesized images in which underestimation of head turns was compensated by adding more head turn while keeping the eyes the same. The error was dramatically decreased for these pictures.
