In order to assess the faceness of a possible candidate face
image, it first needs to be frontal in such a way that the nose
is in the center of image. It is done by applying a translation
and two rotations which map the nose to the center of the
image and maps the other pixels correspondingly. The first
rotation matrix is constructed from unit vectors along the
axes of the rotated space. The first vector u is found using the
two eye points and is along the x-axis of the rotated space.
Taking the cross product of two vectors in the face triangle,
the u vector and the vector between the right eye point and
the nose point, gives a vector w perpendicular to the face
triangle and along the z-axis of the rotated space. The third
vector v is found as the cross product of u and w and is