Analysis of observing and recognition profile facial images using eye tracking system

Abstract

Facial images have been the subject of research for many years, using the eye-tracking system. However, most researchers concentrate on the frontal view of facial images. Much less research has been done on faces shown at different angles or profile views of faces in facial images. However, as we know, in reality we often view faces from different angles and not just from a frontal view. In our research we used a profile presentation of facial images and analyzed memory and recognition depending on the display time and dimensions of the facial images. Two tests were performed, i.e. the observation and the recognition test, and we used the well-known yes/no detection theory. We used four different display times in the observation test (1, 2, 4 and 8 seconds) and two different dimensions of facial images 640 × 480 and 1280 × 960). All facial images were taken from the standardized face database Minear&Park. We measured the recognition success which is mostly presented as a discrimination index A’, incorrect recognition (FA – false alarm) and time-spatial method based on fixation duration and saccade length. In this case, eye tracking provides us with objective results when viewing facial images. In the results it was found that extending the display time of facial images improves recognition performance and that the dependence is logarithmic. At the same time, wrong recognition decreased. Both parameters are independent of the dimensions of the facial images. This fact has been proven by some other researchers also for frontal facial images. It was also discovered that with an increase of the display time of facial images an increase of the fixation duration and saccade lengths occurred. In all results we detected major changes at the display time of four seconds, which we consider as a time, where the subjects looked at the whole face and their gaze returned to the center of the face (in our case eye and mouth).