Information Technology Reference
In-Depth Information
Eye Fundus
Image
Gaussian
pyramid
Locate the
region of the
brightest pixel
(RBP)
Level-0 of
pyramid
Level-N of
pyramid
(…)
Calculate the
RBP centroid
Level-0 of
pyramid
Level-N of
pyramid
(…)
Image subwindow
centered at the
RBP centroid
Fig. 2. Diagram summarizing the image subwindow extraction subtask
Thus, the Laplacian pyramid allows a representation of the original image into
a hierarchy of images in such a way that each level corresponds to a different
band of frequencies.
The characterization of an IP is based mainly on a property by which a point
of the image which belongs to the papillary contour presents a change in its
intensity level in relation to its neighboring pixels. Nevertheless, this property
represents a necessary but not sucient condition, since there may be other
causes that give rise to the appearance of noise associated to other type of
intensity level changes. This noise may be due, for instance, to the non-uniform
color of the retina or to the existence of blood vessels. In order to reduce the
image channel number, we make a weighted sum of the three RGB channels to
produce a single image. Then we select as IPs the brightest pixels in the two
first levels of the Laplacian pyramid applied to this average image. The figure 4
shows an example of the IPs obtained in each of one of these pyramid levels.
2.3 Obtaining Papillary Elliptic Contours by Genetic Algorithms
Genetic algorithms are one of the paradigms most frequently used in evolutionary
computation and they owe their inspiration to the biological process of evolu-
tion: natural selection and survival of the fittest individuals [16]. Thus, given a
problem of a specific domain, these algorithms code potential solutions using a
structure of data like a chromosome (individual) where the genes are parameters
of the problem proposed. The approach to the solution of each chromosome is
calculated using a fitness function. Finally, by applying selection and variation
operators (mutation and recombination), a population of chromosomes evolves
to the optimum solution until a finalization criterion is achieved.
Since the human papillary contour is oval-shaped, this work proposes search-
ing for the solution by approaching this contour with an ellipse, like is done in
other works, e.g. [1,6,15]. From this instant, it is assumed that: (1) the geometric
shape of the papilla can be approached by a non-deformable ellipse, and (2) the
Search WWH ::
Custom Search