Image Processing Reference
In this section, the performance of different c-means algorithms on segmentation of brain
MR images is presented. Details of the experimental set up, data collection, and objective
of the experiments are same as those of Section 2.4.
Consider Fig. 2.3 as an example that represents an MR image (I-20497774) along with
the segmented images obtained using different c-means algorithms. Each image is of size
256×180 with 16 bit gray levels. So, the number of objects in the data set of I-20497774 is
46080. The parameters generated in the discriminant analysis based initialization method
are shown in Table 2.3 only for I-20497774 data set along with the values of input param-
eters. The threshold values for 13 features of the given data set are also reported in this
Table 2.4 depicts the values of DB index, Dunn index, and β index of FCM and
I-20497774: original and segmented images of HCM, FCM, RCM, and RFCM
RFCM for different values of c on the data set of I-20497774, considering w = 0.95 and
m = 2.0. The results reported here with respect to DB and Dunn index confirm that both
FCM and RFCM achieve their best results for c = 4. Also, the value of β index, as expected,
increases with increase in the value of c.
For a particular value of c, the performance of
RFCM is better than that of FCM.
TABLE 2.3 Values of Different Parameters
Finally, Table 2.5 provides the comparative results of different c-means algorithms on
I-20497774 with respect to the values of DB index, Dunn index, and β index. The cor-
responding segmented images along with the original one are presented in Fig. 2.3. The
results reported in Fig. 2.3 and Table 2.5 confirm that the RFCM algorithm produces seg-
mented image more promising than do the conventional c-means algorithms. Some of the
existing algorithms like PCM and FPCM fail to produce multiple segments as they generate
coincident clusters even when they are initialized with final prototypes of the FCM.