Self-Organizing Maps and Their Applications in Image Processing, Information Organization, and Retrieval - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 439

(a)

(b) (c)

FIGURE 11.14
(a) Original image 98640 from the Bridgeman Art Library collection of paintings. (b) Segmented
image by the ICM algorithm. (c) Segmented image by the split-merge SOM algorithm.

the correspondence between the segmented image and the ground truth. The
total rank of the entire segmented image for each case is calculated as

K

1
K

R i = r ij . (11.114)

j = 1

The ranking of the segmented images by the ICM and the split-merge SOM
are presented in Table 11.4. It is seen that the split-merge SOM tends to cap-
ture the ground truth and gives a more detailed segmentation. However, the
ICM is the best for images that contain less details and more homogeneous
regions.