The following images are the templates provided to the algorithm for all the calculations
shown in this report.
Some of the characteristics of these templates are:
Multiple faces are used to capture the different factions and gestures that a
face can have in natural photos. Moreover, all the photos are different
but all of them have face characteristics therefore, if a certain area is
similar to all of them we can be certain that it is a face, while if another area
is by chance similar to only one of them and different to the others, it will still
receive a low correlation value.
All of them show approximately the same area of the face. This is important as
the correlations with the different templates are added and it is important that
they produce the maximum peaks in the same places.
The templates are all looking directly to the camera, that is, the faces are not
turned from the viewers point of view.
Most of the photos are taken with flash. This makes the algorithm more
effective when the photo has been taken with flash due to the common location of shades
and to the similarity of the detected colors.
All the photos are from caucasian subjects, but this has proven to have a negligible
effect in the detection of other races.
Most of the photos are from men, but this does not seem to have any effect on
the algorithm performance with women.
