T. Sziranyi et al., Image segmentation using Markov random field model in fully parallel cellular network architectures, REAL-TIME I, 6(3), 2000, pp. 195-211
Markovian approaches to early vision processes need a huge amount of comput
ing power. These algorithms call usually be implemented on parallel computi
ng structures. Herein, we show that the Markovian labeling approach can be
implemented in fully parallel cellular network architectures, using simple
functions and data representations. This makes possible to implement our mo
del in parallel imaging VLSI chips.
As an example, we have developed a simplified statistical image segmentatio
n algorithm for the Cellular Neural/Nonlinear Networks Universal Machine (C
NN-UM), which is a new image processing tool, containing thousands of cells
with analog dynamics, local memories and processing units. The Modified Me
tropolis Dynamics (MMD) optimization method can be implemented into the raw
analog architecture of the CNN-UM. We can introduce the whole pseudo-stoch
astic segmentation process in the CNN architecture using 8 memories/cell. W
e use simple arithmetic functions (addition, multiplication), equality-test
between neighboring pixels and very simple nonlinear output functions (ste
p, jigsaw). With this architecture, the proposed VLSI CNN chip can execute
a pseudo-stochastic relaxation algorithm of about 100 iterations in about 1
00 CLS.
In the suggested solution the segmentation is unsupervised, where a pixel-l
evel statistical estimation model is used. We have tested different monogri
d and multigrid architectures.
In our CNN-UM model several complex preprocessing steps can be involved, su
ch as texture-classification or anisotropic diffusion. With these preproces
sing steps, our fully parallel cellular system may work as a high-level ima
ge segmentation machine, using only simple functions based on the close-nei
ghborhood of a pixel. (C) 2000 Academic Press.