Contribution to Segmentation of Digital Images Based on Clustering
Jraj Horvath
DepartmentofCybeetics andArtiialIntelligence
FacutyofElecricalEngineeringandInfoatics
TechnicalniversityofKsice
Lea9, 04200Koice,Slovakia
Juraj.Hovath@tuke.sk
Ladislav Madrasz
DeparentofCybeeticsand ticiaIntelligene
FacultyoEleccalEngineeringandInfonatcs
Tecncalniversityo Kice
Letn 9, 04200 Kosice,Slovaka
Ladislav.Madarasz@tuke.sk
A- In this work are described methods of segmentation
of digital images based on clustering. As clustering method
was chosen fuzzy c-means clustering method, but these
methods of segmentation are not restricted only to fuzzy c
means. In these methods of segmentation can be used any
clustering method, which respects some rules. Colour
perception by humans is also important part of this
contribution. s a logical consequence of colour perception is
L*u*v* colour ,which is also deined and described. It is
described L*u*v* transformation and application of this
transformation in image segmentation. First of all is
mathematical definition of partial segmentation required for
introduced segmentations' methods.
I.INTRODCTION
Imagesegmentationwas,is andwillbamaor research
toc or man image roessing researhers. he reasons
areobviousandapplicationscountendless. Mostcomputer
vsion and imageanasis proemsreuire a sementaion
stage in order o detect oects or divide the image into
regions, whch can be considered homogeneous according
toa given criterion, sucha colour, motion, texture, etc [2,
3
.
Sometime is necessary to adust computer vision to
human vison. Especially is it necessa, when we are
sementing images. which were semented by people and
we y to repace people with computers o when we want
o he people n semenation of iages. Typical
appicaion s meicine, e.g. sementation oI images
ordenatological images 2,3.
II.PARTIALSEGENAIONDEFINITION
For claim of methods is necessay to express more
mathematically exactly artial sementation.which is not
exactlymathematicallydenedin2.3, 10.
Let I mars digtal image of rectangular shape. Let
image 1 has wid w and height h. Let
a E I, = {1,2, ... , w}, j= 1,2, . . . , h} marks pixel of image
I(in general pixel can be a vectorof vaues). Let R; I
0-7803-8588-8/4/$20.00 24EEE.
Iveta Zolotova
Deparent of CybeticsanAicial Intlligence
Facultyof Elecical Engineering andInfoatics
TechnicalniversityofKosice
Lea 9, 04200Koice,Slovakia
Iveta.Zolotova@tuke.sk
marks segment, let segment's numer is M. Let H marks
homogenous criterion, let homoenous criterion is biny
(itcanreachonytwovalues)4, 10
{ I> argumetaccomplishriterion
H=
argumennot accomplishcriterion
(I)
Digita images segmentation is process f dividing
imageI tonotoverappingsemens R;. Eachement R;
accompish homogenous crierion H and at the same time
or each neighong sements R applis, that by
homogenous criteron uniting nghbong seent Rj
and sement R; wilecreaed not homogenous sement.
Semnt R, i part o imae I, which was created in
process o seentaion. Dgitl image segmentation is
process, which rest accomplish next oitions 2, 3, 4,
10
89
;=1
V n,m E {I, 2, . . . , M}, n'm : R Ó R = 0
V n E {I. 2, ... , M}: H(
R
n) = 1
n.m E {I, 2 .. ... M}, n,m,R and R,
whicareebours:H(Rn U R ) = 0
III. ANS COLORPERCEPTION
(2
)
(3)
(4
)
(5)
n human ee are hree ierent sensors for colour
perception. These receptors are called cones. Eveyone
om these receptors hae dierent visal pient, which
respods on photos. One pigment has mimm
sensiliy on waeleng of 455 nometres (blue
pment), seononwavelen o535 nanomees (een
pigment) and third has maximum on wavelength of 570
nomeres(redpient).
Curvescorrespondedo sensibilityofeachpimentwere
gotexperimentalandweregotom[8.