Design and Implementation - Segmentation of Hand Bone for Bone Age Assessment - page 82

terminated, whereas, if the ratio is in the moderate range, then the CER results in
high value, and the dividing process continues.
The procedures of mentioned process can be viewed as constructing a decision
tree for classifying blocks that should continue undergoing quadruple dividing and
blocks that should stop undergoing quadruple dividing process. This kind of deci-
sion tree outperforms other sophisticated cluster validity in this context in terms
of computational demand. Beside, this kind of decision flow generates compre-
hensive rules for users to understand and manipulate. The main component in this
decision flow has been defined. The entire process can be better illustrated via a
decision flowchart shown in Fig. 3.12 .
As we can observe in Fig. 3.12 , there are many vague and ambiguous adjec-
tives such as 'close to', 'far apart', 'high', 'low', 'dark', 'bright' 'moderate',
'medium', 'less likely', 'most likely' and 'least likely'. These words are common
terms in daily linguistic communication to describe intuitive a priori knowledge.
In terms of semantics, these words carry no precise definition and are very vague
to be measured in exact numerical number. First of all, the question is that why
we describe the relations of the cluster in vague words instead of exact numerical
numbers, for example, if the NCL , below 0.3 or any numerical threshold, then it is
considered as 'near 0'? Basically, there are three reasons:
1. To fulfil the important criterion of P8 in which we desire to impose human vis-
ual perception intuition into this automated scheme so that it behaves auton-
omously while still manage to resemble human decision making process and
cognitive ability; human do not use numerical numbers, instead, human utilize
intuition and a priori knowledge to perceive and interpret an image, and yet we
human beings are capable of classifying extremely complicated objects from a
very chaotic image. In the case of NCL , it is not logical for human to interpret
clusters that produce NCL below 0.3 are considered as 'near 0', and NCL value
such as 0.31 is considered as not close to 0. The cognitive process of human
simply do not behave this way and thus any pre-set thresholding is not effec-
tive in imitating human cognitive visual ability. Moreover, pre-set thresholds,
as mentioned in previous chapter, is not adaptive to the block's content and thus
does not fulfil criterion P10 in handling variability.
2. The cause and effect are investigated through empirical study and human obser-
vation; the analytic function and precise numerical relation between the cause
and effect in the complex system containing pixel intensity clusters, optimum
block size, resultant segmented image, are not yet available. When these ana-
lytic function and precise definition do not exist, only vague linguistic seman-
tics can model the human observation in the complex system.
3. The exact numerical solution or threshold is not necessarily required to perform
crucial decision in the complex system. Decision making using approximation gain
significant computational efficacy advantage that is paramount in fulfiling criterion
P2 which is a principal advantage of the proposed scheme compared to other com-
putationally demanding scheme such as active shape model and active appearance
model. Therefore, if approximations involve, vague descriptive terms are inevitable.