In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
wrt some kind of reference, so that subject’s transformed model 
appears to be staring at the camera. This problem exposes the 
following six degrees of freedom: three angles of rotation 
(referred to as pitch, yaw and roll) and three measures of 
bounding box translation (xc, ye, zc), respectively relative to 
the three main axes (X, Y and Z), as shown in Figure 2: 
Figure 2. Main axes and corresponding rotation angles 
representation. 
The whole algorithm is conceived as a two-step optimization 
process. In the first phase we exploit natural vertical symmetry 
of human face, acquiring knowledge about the yaw and roll 
angles, along with the xc translation parameter. We designed a 
simple measure of the asymmetry between the two halves of a 
human head defined by intersection of its 3D model with a 
hypothetical symmetry plane, and called it Mean Symmetry 
Distance (MSD). It is obtained by projecting a bundle of 
parallel lines, perpendicular to the given plane, and intersecting 
it with the triangle mesh of the head. Then, for each line in the 
bundle, intersecting points closest to the plane are selected by 
each side (two total points) and the absolute difference between 
the distances of these points from the plane itself is computed, 
if applicable (that is, if the projected line actually intersects 
each part of the head at least once). Finally, the mean of these 
absolute differences will be assumed as a symmetry error value 
(MSD) for the considered plane. The density of the line bundle 
may vary according to desired accuracy. In the second phase we 
lean on natural vertical development of the central profile shape 
of human faces, gaining clues about the pitch angle. First, a 
sampling of the profile is extracted, resulting in a quasi- 
continuous vector of points, then the parameters of a first order 
equation are adapted for the best possible approximation of this 
vector. In other words, we try to find the line that best fits the 
shape of the profile, describing it by its explicit slope-intercept 
parameters. The error measure between the profile and its 
estimate, named Mean Profile Distance (MPD), is straight 
forwardly computed as the mean of point-line distances 
between the given linear equation and the sampling vector. 
tj 
Figure 3. Extracted frontal profile shape and possible 
approximating lines, with the right one intuitively having a 
heavily lower mpd value. 
Figure 4. Head 3D model before (left) and after (right) MSD 
minimization process. 
3. 3D FACE ENROLLMENT AND RECOGNITION 
For the acquisitions of “.ase” files has been used the 
MainAxis’s “3D CAM SURF ACER”, a 3D laser scanner based 
on structured light with a resolution of 640 by 480, which is a 
good resolution for 3D processing. 3D acquisitions can be 
affected by random noise that consists of spikes and holes 
(Bowyer, K., Chang, K., Flynn, P., 2006), so has been 
necessary a pre-processing phase. The presented technique of 
3D vectorial photography is a potentially low cost and has 
solid-state robustness. It can be used as an alternative for other 
scanning or multi-image techniques. For distances ranging 
froml tolOmiters the same order precision with the high 
acquisition rate. Furthermore, it has no moving parts(excluding 
light source cool fan),and uses white light illumination. In spite 
of two-acquisition based principles it permits registration, with 
a normal speed camera, of 3D images of objects moving with 
the velocity up to 1 m/sec.in darkened laboratory and up 
to20mm/sec.in normal ambient illumination conditions. 
3.1 Individuation of nasal pyramid points 
We have implemented the algorithms to individuate the point of 
the nasal pyramid.