These
two pictures are taken from the works of the Russian psycho-physicist Yarbus,
dating back to the 1950s. The picture on the right shows the trace of the
gaze of a subject exploring the portrait on the left. Yarbus demonstrated
that human beings, as these pictures show us, do not scan a scene in a
raster-like fashion. They rather perform jumps,
known as saccades, between the different points of interest, on
which fixation is maintained for a short period.
We do not use saccades to paint a complete internal representation of a visual scene. A few experiments suggest that we rather rely on the external word for storing information and only remember pointers to relevant locations in the scene. We then make use of saccades to retrieve the information as we need it. Therefore, saccades constitute a way to select task relevant information. This is confirmed by the fact that, as Yarbus already noticed, the saccadic pattern depends on the cognitive task to be performed.
In these images we remark that most of the time is spent looking at the eyes and the mouth. Other studies show that these are the regions we mostly rely on for face recognition. We therefore focus on those regions for our Facial Features detection and Face Authentication algorithms.