Meta-learning for Image Processing Based on Case-Based Reasoning - Computational Intelligence in Healthcare: Advanced Methodologies - page 239

Fig. 25. Dendrogram for image description based on weighted texture, centroid and statistical
features

(a) Image neu4 (b) Image neu4 rotated 180

degree

(c) Result for neu4 with
parameter-set of neu4_r180

Fig. 26. Orignal image and rotated image and influence of the parameter set

To sort out rotated images from the group including the un-rotated images, we
have to give more emphasis to the feature centroid, because this feature is the only
one which is not invariant for rotations. Therefore, we divide the image features set
into three groups: texture features, centroid, and the remaining statistical features.
Each group gets a total weight ω g of 1/3. The local weights ω gi in each group are

computed as follow

l l

1 1

∑ ∑

ω = ω = =

,

(8)

g gi

3 * l 3

i = 1 i = 1

where l is the number of features in the group.