Rough-Fuzzy Clustering Algorithm for Segmentation of Brain MR Images - Rough Fuzzy Image Analysis: Foundations and Methodologies - page 30

TABLE 2.4 Performance of FCM and RFCM on
I-20497774 data set
Value DBIndex DunnIndex β Index
of c FCMRFCM FCMRFCM FCM RFCM
2 0.38 0.19 2.17 3.43 3.62 4.23
3 0.22 0.16 1.20 1.78 7.04 7.64
4 0.15 0.13 1.54 1.80 11.16 13.01
5 0.29 0.19 0.95 1.04 11.88 14.83
6 0.24 0.23 0.98 1.11 19.15 19.59
7 0.23 0.21 1.07 0.86 24.07 27.80
8 0.31 0.21 0.46 0.95 29.00 33.02
9 0.30 0.24 0.73 0.74 35.06 40.07
10 0.30 0.22 0.81 0.29 41.12 44.27

TABLE 2.5 Performance of Different C-Means
on I-20497774 data set
Algorithms DBIndex DunnIndex β Index
HCM 0.17 1.28 10.57
FCM 0.15 1.54 11.16
RCM 0.16 1.56 11.19
RFCM 0.13 1.80 13.01

TABLE 2.6 Haralick's and Proposed Features on I-20497774 data set
Algorithms Features DBIndex DunnIndex Index Time(ms)

HCM H-13 0.19 1.28 10.57 4308
H-10 0.19 1.28 10.57 3845
P-2 0.18 1.28 10.57 1867
H-10

β

P-2 0.17 1.28 10.57 3882
FCM H-13 0.15 1.51 10.84 36711
H-10 0.15 1.51 10.84 34251
P-2 0.15 1.51 11.03 14622
H-10

∪

P-2 0.15 1.54 11.16 43109
RCM H-13 0.19 1.52 11.12 5204
H-10 0.19 1.52 11.12 5012
P-2 0.17 1.51 11.02 1497
H-10

∪

P-2 0.16 1.56 11.19 7618
RFCM H-13 0.13 1.76 12.57 15705
H-10 0.13 1.76 12.57 15414
P-2 0.13 1.77 12.88 6866
H-10∪P-2 0.13 1.80 13.01 17084

∪

2.6.1 Haralick's Features Versus Proposed Features
Table 2.6 presents the comparative results of different c-means for Haralick's features and
features proposed in (Maji and Pal, 2008) on I-20497774 data set. While P-2 and H-13 stand
for the set of two proposed features (Maji and Pal, 2008) and thirteen Haralick's features,
H-10 represents that of ten Haralick's features which are used in the current study. The
proposed features are found as important as Haralick's ten features for clustering based
segmentation of brain MR images. The set of 13 features, comprising of gray value, two
proposed features, and ten Haralick's features, improves the performance of all c-means
with respect to DB, Dunn, and β. It is also observed that the Haralick's three features -
sum of squares, information measure of correlation 1, and correlation 2, do not contribute
any extra information for segmentation of brain MR images.

2.6.2 Random Versus Discriminant Analysis Based Initialization
Table 2.7 provides comparative results of different c-means algorithms with random initial-
ization of centroids and the discriminant analysis based initialization method described in