Pit Pattern Classification in Colonoscopy using Wavelets - WaveLab

4.2 Classification based on features
Energy
∑w−1
∑w−1
f i = C i (x, y) 2 (4.16)
y=0 x=0
Entropy
∑w−1
∑w−1
f i = C i (x, y) log ∗ (C i (x, y)) (4.17)
y=0 x=0
Inverse difference moment
∑w−1
∑w−1
f i =
y=0 x=0
C i (x, y)
1 + (x − y) 2 (4.18)
Cluster shade
∑w−1
∑w−1
f i = (x − m x + y − m y ) 3 C i (x, y) (4.19)
y=0 x=0
with
∑w−1
∑w−1
P x (i) = C(i, j) P y (j) = C(i, j) (4.20)
j=0
i=0
and
∑w−1
∑w−1
m x = iP x (i) m y = jP y (j) (4.21)
i=0
j=0
Cluster prominence
∑w−1
∑w−1
f i = (x − m x + y − m y ) 4 C i (x, y) (4.22)
y=0 x=0
When all feature vectors F i for all images in I T have been calculated, an arbitrary image I α
can be classified by using a classifier such as k-NN, LDA, CART or support vector machines
(SVM). Throughout this thesis however the only classifiers used are k-NN and SVM. The
classification process based on the features just calculated and the according classifiers are
described in section 4.4.
41