Satellite Image Classification Using Cascaded Architecture of Neural Fuzzy Network - Advanced Techniques in Knowledge Discovery and Data Mining - page 215

down the next row. Then the window will move in a zigzag pattern until the entire
image has been covered. As an illustrative example (Fig. 8.4), the first block in the
top-left corner is the first co-occurrence matrix. Then the label S on this block is
the last column to be subtracted while the label A on the next matrix is to be added.
We repeat this process column by column and row by row in a zigzag pattern.
Because the data are coded in zigzag format, they should be regularized after
the process finished. Although it is not easy to design the SLL, it is the fastest
method to calculate the co-occurrence matrix with full color range. Moreover, this
update method can be applied to other applications.

Fig. 8.3. The equal probability quantizing (EPQ) algorithm.

S

S

S A

. . .

S

A

A S

. . .

A

Fig. 8.4. The symmetric linked list (SLL) algorithm.

8.2.2.3 Global Texture Features from Second-Order Histogram Statistics

From the co-occurrence matrix, we can define the features as follows:
z Entropy (ENT). The entropy computed from the second-order histogram pro-
vides a standard measurement of homogeneity and is defined as

¦¦

Entropy p ( i , j ) log( p ( i , j )).

(8.2)

i j

Higher values of homogeneity will indicate fewer structural variations whereas
lower values will be interpreted as a higher probability of textural region.
z Contrast (CON). The contrast feature is a difference moment of the P matrix
and a standard measurement of the number of local variations presented in an
image. The contrast feature on the second-order histogram is defined as