T. Sziranyi et J. Zerubia, MARKOV RANDOM-FIELD IMAGE SEGMENTATION USING CELLULAR NEURAL-NETWORK, IEEE transactions on circuits and systems. 1, Fundamental theory andapplications, 44(1), 1997, pp. 86-89
Markovian approaches to early vision processes need a huge amount of c
omputing power. These algorithms can usually be implemented on paralle
l computing structures. with the Cellular Neural Networks (CNN), a new
image processing toot is coming into consideration, Its VLSI implemen
tation takes place on a single analog chip containing several thousand
s of cells. Herein se use the CNN UM architecture for statistical imag
e segmentation. The Modified Metropolis Dynamics (MMD) method can be i
mplemented into the raw analog architecture of the CNN, We are able to
implement a (pseudo) random held generator using one layer (one memor
y/cell) of the CNN. We can introduce the whole pseudostochastic segmen
tation process in the CNN architecture using 8 memories/cell, We use s
imple arithmetic functions (addition, multiplication), equality-test b
etween neighboring pixels and very simple nonlinear output functions (
step, jigsaw, With this architecture, a real VLSI CNN chip can execute
a pseudostochastic relaxation algorithm of about 100 iterations in ab
out 1 ms, In the proposed solution the segmentation is unsupervised. W
e have developed a pixel-level statistical estimation model. The CNN t
urns the original image into a smooth one. Then we have two gray-level
values for every pixel: the original and the smoothed one. These two
values are used for estimating the probability distribution of region
label at a given pixel. Using the conventional first-order Markov Rand
om Field (MRF) model, some misclassification errors remained at the re
gion boundaries, because of the estimation difficulties in case of low
SNR. By using a greater neighborhood, this problem has been avoided.
In our CNN experiments, we used a simulation system with a fixed-point
integer precision of 16 bits, Our results show that even in the case
of the very constrained conditions of value-representations (the inter
val is (-64, +64), the accuracy is 0.002) can result in an effective a
nd acceptable segmentation.