This results in very few candidates and ideally only one as
it is shown in (Figure 3(d)). Regardless of the number of
the found candidates, they will be all registered to a frontal
face, and then their faceness will be measured using their
depth information.