(XIX-B3, 2012
SENSING
sion uncertainty
neters. This paper
onent *m' as well
le parameters for
riterion has been
1ese criterions are
based on outputs
opy is sensitive to
ly. Soft classified
ing SCM, overall
accuracies, while
iges for different
ich gives highest
7 noise clustering
ta collection and
any classification
solution to this
ild be one where
1d removed from
introduced such
noise class. The
ctional type (K-
ability to detect
tly demonstrated
e to both fuzzy
regression based
plays a pivotal
use without the
eparate the good
ccess the quality
istering problem
fore it computes
m, 1997).
ut entropy is not
/ set theory and
'ompare several
ew of robustness
this paper is to
of FCM, PCM
espect to fuzzy
' as uncertainty
; of parameters
without entropy
SESSMENT
iced by Bezdek
rique each data
point belongs to a cluster to some degree that is specified by a
membership grade, and the sum of the memberships for each
pixel must be unity. This can be achieved by minimizing the
generalized least - square error objective function in Eq. (3),
Ni c = 2 3
JU.) -EX(u) |x, =x, (3)
Subject to constraints Eq (4),
Yu, 21 for alli
N
LH, >0 for all j @
Osi <1 for all i, j
where X; is the vector denoting spectral response of a pixel i, x
is the collection of vector of cluster centers x,, pj are class
membership values of a pixel, c and N are number of clusters
and pixels respectively, m is a weighting exponent (1<m<co),
which controls the degree of fuzziness, | X x is the squared
distance (di) between X; and x,, and is given in Eq (5),
à -|x,-x[, -(X,-x,) 4(x, -x;) (5)
where A is the weight matrix. Amongst a number of A-norms,
three namely Euclidean, Diagonal and Mahalonobis norm, each
induced by specific weight matrix, are widely used. The
formulations of each norm are given as (Bezdek, 1981) in Eq
6),
= I | Euclidean Norm
Az D Diagonal Norm (6)
Az c Mahalonbis Norm
Where I is the identity matrix, D; is the diagonal matrix having
diagonal elements as the eigen values of the variance covariance
matrix, C; given in Eq (7),
c Xxx) y e
The class membership matrix pj is obtained in Eq (8) and (9);
1
c ( q? Von (8)
s)
k=1
j
where di-Y? (9)
j=l
2.2 Possibilistic c-Means Approach (PCM)
In PCM, for a good classification is it expected that actual
feature classes will have high membership value, while
unrepresentative features will have low membership values
(Krishnapuram and Keller, 1993). The objective function,
which satisfies this requirement, may be formulated in Eq (10);
N c » 2 c N o. (10)
4U)- 3 Yu) X, -v| +21, 2 (1-4)
Subject to constraints;
max ,u, > 0 foralli
N
SN 20 for all j
isl
Osu, si for all i, j
uj is calculated from Eq. (8).
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
In Eq. (10) where n; is the suitable positive number, first term
demands that the distances from the feature vectors to the
prototypes be as low as possible, whereas the second term
forces the ju; to be as large as possible, thus avoiding the trivial
solution. Generally, n; depends on the shape and average size of
the cluster j and its value may be computed in Eq (11);
S 2
> upd
n, K-—— (11)
m
2,
i=l
Where K is a constant and is generally kept as one. After this,
class memberships, uj are obtained in Eq (12);
zu io y (12)
y d? Ao
1+| —
N;
2.3 Noise clustering without Entropy
A concept of "Noise Cluster' was introduced such that noisy
data points may be assigned to the noise class. The approach is
developed for objective functional type (K-means or fuzzy K-
means) algorithms, and its ability to detect 'good' clusters
amongst noisy data is demonstrated (Dave and Krishnapuram,
1997). Noise clustering, as a robust clustering method,
performs partitioning of data sets reducing errors caused by
outliers. In many situations outliers contain important
information and their correct identification are crucial. NC is a
method, which can be adapted to any prototype-based
clustering algorithm like k-means and fuzzy c-means (FCM)
(Frank, et al. 2007). The main concept of the NC algorithm is
the introduction of a single noise cluster that will hopefully
contain all noise data points . Data points whose distances to all
clusters exceed a certain threshold are considered as outlier.
This distance is called the noise distance. The presence of the
noise , cluster allows outliers to have arbitrarily small
memberships in good clusters (Dave, et al. 2007). In other
classifiers where noise data points are not separate and present
in information class, may lead to some kind of information
scepticism. The objective function, which satisfies this
requirement, may be formulated in EQ (13), (14) and (15);
U (u, DEP 9 (13)
i
c d; Hom) d Jm) (14)
i > ; HR
k=1 di ö
Where I=k=¢
Where 1=j=c¢
and
1 -l
e S Jom) (15)
Hon = = TS +1
j=1 d;
O> 0, any float value greater than zero
Where 8 >m>1, (any constant float value more than 1)
N- row * column (image size)
i = stands for pixel position at i^ location distance between X;
and Vj