Eye tracking study of frontal and profile face image observation and recognition

Abstract

Facial images are an important element of nonverbal communication. Eye-tracking systems enable us to objectively measure and analyse the way we look at facial images and thus to study the behaviour of observers. Different ways of looking at facial images influence the process of remembering faces and recognition performance. In the real world we are dealing with different representations of faces, especially when we look at them from different angles. Memory and recognition performance are different when test subjects look at the face from the frontal or from a profile view. We studied crossobservation and recognition, so we performed two tests. In the first test, subjects observed facial images shown in the frontal view and recognized them in the profile view. In the second test, the faces were observed from the profile and recognized in the frontal view. The presentation time in the observation test was four seconds, which was found to be an adequate time for sufficient recognition in some previous tests. The results were analysed with the well-known time and spatial method based on fixations and saccades and with the new area method using heatmaps of the eye tracking results. We found that the recognition success (correct and incorrect recognition) was better when the combination of frontal view and profile recognition was used. The results were then confirmed by measuring the fixation duration and saccade length. More visible facial features resulted in a shorter fixation duration and shorter saccade length, which led to a better memory. We also confirmed the results of observation and recognition by area analysis, where we measured the area, perimeter and circularity of heatmaps. Here we found that larger areas and perimeter and smaller circularity of heatmaps resulted in better memory of facial images and therefore better recognition.