pletely sto-
he usual a
(x, y) pixel
e algorithm
of segmen-
te. To pro-
in probabil-
to obtain a
om estima-
(- mentioned
hms can be
n.^V.«]99].
from image
ete Markov
for Vi zit;
on number;
egion types;
n.
vergence 1s
ity of (x, y)
! brightness
Xels;, ati the
S ecnormali-
Zzation factor and. 7(/) must tend to infinity with an
unlimited Zinerease: of. à but no quicker than
T(I) € kIn(j -1): k - a parameter. depending on spe-
cilic characteristics of two-level Markov random
field.
If the above conditions are met, then the random
estimation of S, pixel state in the limit with an un-
limited increase of i transforms into an ordinary
estimation by a posteriori probability maximum cri-
terion and random estimation sequence of pixels
states
yu ya tr Ny xy vee
is a nonstationary Markovian chain of the first order
with a single absorbing state corresponding to the
global optimum.
Thus. during initial iterations the algorithm searches
for global maximum regions compensating the low
quality of the initial segmentation and then tends to
a deterministic form and provides the optimal image
segmentation.
MODELLING RESULTS
Figs 2-5 illustrate the implementation of algorithms
with random decision rules. The algorithm described
in (Lisitsyn. V., 1991) was considered as determinis-
tic algorithm. Fig. 2 shows reference synthesized
laser locator image with superimposed speckled
noise. Fig. 3 shows the result of initial segmentation.
Fig. 4 presents the result of deterministic segmenta-
tion algorithms implementation. Fig. 5 shows the
result of segmentation algorithms implementation
with random decision rules.
Mathematical modelling results have shown that in
case of proper initial segmentation the stochastic
and deterministic algorithms implementation yields
equal results while in case of improper initial seg-
mentation, the resulting image provided by stochas-
tic algorithm implementation much better agrees
with the ideal segmentation. Besides, algorithm im-
plementation time increases by 30-35%.
The proposed approach is of general nature and can
be applied not only to laser locator image segmenta-
tion but also to other similar cases.
REFERENCES
Borisenco. 1.. Zlatopolsky, A., Muchnic, A.. 1987. Seg-
mentaciya izobrajeniy (Image Segmentation). Avtomatica i
telemechanika, N 7. pp. 4-25.
Dansac. L. Meyzonnetie. 1.1985, CO» -laser- doppler
rangefinding with heterodvne detection and chirp pulse
compression. Proceedings of the SPIE. V. 590. pp.380- 388.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Derin. H.. Cole. W.S., 1986. Segmentation of textured
images | using Gibbs randomfield. | Computer vision,
graphics and image processing. V. 35. pp.72-98.
Duda, R.O.. Hart, P.F.,1973. Pattern classification and
scene analysis. John Wiley, New York.
Hanson, F.R., Elliott, H., 1982. Image segmentation using
simple Markov fields models . Computer vision, graphics
and image processing, V. 20, pp.101-132.
Kelly, P.A.. Derin. H.. Hartt, K.D.. 1988. Adaptive seg-
mentation of speckled images using a hierarchical ran-
dom field model. IEEE Transactions on Acoustics. speech,
and signal processing. V. 36 (10)
Lisitsyn, V.. Obrosov, K., Pasechny, N., .Stefanov, V.,
1990. Laser radar images iterative segmentation for dy-
namic scenes. In: Archives of ISPRS, Victoria, Canada,
Vol. XXVIII, Part 7-1, pp. 716-723.
Lisitsyn, V.^Obrosov, K., Pasechny, N.,' Stefanov, V.,
1991. Doppler optical image segmentation algorithm. So-
viet journal of computer and systems sciences, V. 29 (l1),
pp. 33-42.
Lisitsyn, V., Obrosov, K., Pasechny, N., Stefanov, V.,
1995. Iterative algorithm for segmenting multidimensional
images. Journal of computer and systems sciences interna-
tional, V. 33 (5), pp. 89-98.
Therrien..C.W., Quatiere, T.F.. Dadjion, T.E., 1986. Statis-
tical model-based algorithm for image analysis. Proceeding
of the IEEE. V.74 (4), pp. 532—551.
Wang, ..Y.,1984. Imaging CO? - laser radar field test. Ap-
plied opties. V. 23 (15), pp. 2565-2571.
