TELKOMNIKA
Genetic Optimization of Ne algorithm that transforms a se
operations based on evolution be used, and after having ap
among individuals that stand one generation to another a s
operators are selection, cross being a chain of characters l
chains of chromosomes, and reflects the fitness.
There exists a divers voting method, fuzzy integrat
section for illustrative purpose Integration by Gating
tasks learned through the mo Gating Network are: best ove
classifiers, need not be the sa There are several im
important is by nature of using neuron to evaluate the perform
gating network is based on a networks of experts [5]. In Figu
Figure 1. Repre
2.1. Iris Image Preprocessin
Due to the unique, identification based on the iri
The idea of using iris patt ophthalmologist Frank Burch.
Safir, American ophthalmolog system, led to contact with Jo
he developed the necessary a [7]. These algorithms, patente
of all iris recognition systems
Various studies carrie which uses neural networks a
identifier is perhaps one of t appearance of the iris. This ide
The database of huma China [58]. This institution has
the database, which consists o people the total database. Th
8 images were used for trainin ISSN: 1693-6930
Neural Network for Person Recognition Based on . set of mathematical individual objects with regard
ion. The Darwinian laws of reproduction and surviva appeared of natural form a series of genetic op
nd out for the sexual recombination [25], [26]. For a series of genetic operators are applied. The mos
ssover and mutation [15]. Each of the individuals letters or numbers of fixed length that adjust to
d one associates to them with a certain mathema ersity of methods of integration or aggregation of
ration, and gating networks [25]. However, we co ses on the gating network method.
ng Network: in this case decomposition of a learn modules of cooperation is performed. The benefit
verall performance, reuse of existing patterns hete same type; different features can be used for differe
implementations of the modular neural networ ing the gating network. In some cases, this corres
ormance of the other modules of experts. Other em a neural network trained with a different data se
igure 1 a scheme of the gating network integrator is
presentation of the gating network integration meth
ing
, stable and accessible characteristics of iris p iris pattern has become one of the most reliable
atterns to identify people was first proposed ch. However, it was not until 1987, when Leonard
ogists, patented the concept of Burch. His interest John G. Daugman, then a professor at the Univers
y algorithms for biometric recognition through the nted by Daugman in 1994 and partly published in [
that exist today. rried out for iris recognition, as the work of M. Ah
and the cosine transform for iris-based identificati f the most foreign to people, as among us do n
identifier is one of the most accurate among biome man iris is from the Automation Institute of the Acad
has several databases of iris, and we used in this ts of 14 images per person 7 of each eye, we use
The image dimensions are 320x280 pixels, the form ning and 6 for testing.
.… Patricia Melin 311
rd to the time using ival of the fittest can
operations between or the passage from
ost commonly used ls is in the habit of
to the model of the atical function that
of information, like concentrate in this
rning task into sub fits of working with
eterogeneity expert erent classifiers.
ork, but the most esponds to a single
embodiment of the set for training the
r is presented.
thod
patterns, personal le techniques [1-4].
in 1936 by the ard Flom and Aran
st in developing the ersity of Harvard so
e pattern of the iris n [14], are the basis
Ahmad Sarhan [7], ation. The iris as an
not recognize the etric systems [7].
cademy Sciences of is work version 3 of
sed only the first 77 ormat is JPEG, and
TELKOMNIKA
Vol. 10, No. 2 312
In the pre-processing noise removal, in order to extr
filters on it, this in order to he images see Figure 2.
Figure 2. Gene Figure 3 shows the re
maximum and minimum. As s outer parts of the circle, leavin
of the iris, and the other way the image for the network.
Figure 3. Result of ap Once we have all the
the modular neural network, w transform 2D rate with “symml
We proceed to vecto another array of vectors with
67-99. 2, June 2012 : 309 – 320
ng stage, different methods were applied for featu xtract the region of interest iris of the captured im
help the modular neural network, to obtain a high
neral diagram of pre-processing for the CASIA data resulting image, after making a cut to the image,
s shown in the image this can be done in 2 ways: ving them in black and giving us a better apprecia
is to let the image with their property and leaving
applying different techniques of vision to the center he database with pre-processing and before putting
, we compressed the images to 320 x 280 25x25 mlet” of order 8, with 2 levels of decomposition see
torize each image within a matrix containing the h the following 33 34-66, and the last 33 persons
ISSN: 1693-6930 ture extraction and
image, apply some h recognition of the
tabase e, according to the
ys: one is filling the ciation of the center
ing more features in
er of the iris ing each image into
25 using a wavelet see Figure 4.
he first 33 persons, ns in the same way
TELKOMNIKA
Genetic Optimization of Ne Figure
In the first 33 persons samples of the right eye and
training and 6 images to valid then you have two arrays of v
validation matrix, with 8 33 shown in the validation matrix
the modules of the modular ne
2.2. Statement of the Proble