t 100
(2)
f the
nces,
f the
hase
ed in
PD1
n AL.
llates
ongth
an be
llows
stem
| one
ution
ii) an
eam
er at
arm
uring
95
363
3. SPECIAL FEATURES AIMED AT IMPROVING THE SYSTEM PERFORMANCES
Particular care has been devoted to enhance the flexibility of the whole field profilometer. Firstly, a liquid crystal display
projector substitutes the fixed-grating (slide) projector, and allows adaptation of the spatial period of the grating
according to the geometry of the object of interest. Secondly, suitable phase demodulation has been developed which
(i) allows the system to manage both coarse and fine changes of the grating period, and (ii) reduces the computation
time (Sansoni et al, 1994a). Thirdly, a purposely designed matching procedure has been implemented to obtain a
good resolution of the measurement even in the presence of discontinuities of the object shape: in fact the projection of
a coarse grating allows us to evaluate the measuring range, and the projection of a finer grating allows us to achieve a
good resolution in correspondence with fine surface details (Sansoni et al, 1994b). Finally, suitable image
preprocessing has been performed in order to (i) decrease the influence of background illumination and of
nonuniformities of the object reflectivity on the measurement, and (ii) enhance the fringe contrast in the presence of low
diffusive surfaces (Biancardi et al., 1994).
3.1. Image preprocessing
Both linear and nonlinear algorithms have been developed for image preprocessing. Linear preprocessing is based on
image background subtraction and normalization of the projected grating to the illumination intensity. It is implemented
on a pixel by pixel basis, and runs in parallel with the image acquisition (Sansoni et al., 1994b). This procedure has
been verified to work satisfactorily in a wide range of situations. A typical example is shown in Fig. 4.a, which presents
the fringe pattern deformed by a trapezoid shaped object, which is characterized by a highly reflective surface. In this
measurement, a grating of period p=12.5 mm is projected on the target. The illuminated field is equal to 786x528 mm.
The level of background illumination is kept very low due to the non uniform, high reflectivity of the object. Fig. 4.b
shows the effect of the linear preprocessing, which allows us to precisely reconstruct the deformed pattern.
n
RTT
Ls
4 D
2
nt
€ poetas
PAAR EA Tee rd)
oa 1
CON AANA LA AN I I a
Fig. 4. Example of the performance of linear preprocessing on a reflective surface. 4.a: Image of the
pattern deformed by the object; 4.b: Image after the elaboration.
Nonlinear preprocessing is based on a suitable combination of Laplacian directional filters and quadratic filters aimed
at enhancing the contrast of the fringes without enhancing the noise of the image. The method proposed may be
viewed as a Nonlinear Unsharp Masking, where the sharpening signal is produced by a nonlinear filter which detects
correlated elements; the correlation measure allows us to discern details from noise. The orientation of an image detail
is determined by making use of directional sub-operators.
The implemented algorithm has been verified to perform satisfactorily in the presence of very low-contrast, high noise
images. This is the case of the projection of fringes on low-diffusivity surfaces. A typical example is shown in Fig. 5.a.
The object, representing a human face, is made of frosted glass. In this experiment, a grating of period p=4.51 mm is
projected on the target. The illuminated area is 167x118 mm. The linear preprocessing works satisfactorily only in
correspondence with the fringes projected on the flat surface on which the object is placed, whereas the grating
deformed by the object has very low contrast and a high noise. Fig. 5.b shows the effect of the nonlinear procedure: a
marked increase of the fringe contrast is evident on the object, even in the regions where the surface presents steep
slope changes.
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995