Image Processing Reference
In-Depth Information
In the experiments, four independent algorithm-based populations have been created listed
in the subsequent paragraphs. For each population, 300 iterations have been gathered.
In case of CCD RECA, FCD RECA, FFT RECA and FFD RECA 5 populations have
been created. The selection of RECA algorithms have been motivated by requirement of
obtaining solutions with high values of rough fuzzy measures.
Population CCD - CCD RECA The algorithm has been run 5 times, every time pop-
ulation size was 30 solutions that were iterated 300 times for each population. Distance
threshold
has been set to 50.0. The details have been given in Table 11.1. The algorithm
flow as described in Algorithm 2.
Population FCD - FCD RECA The algorithm algorithm has been run 5 times, every
time population size was 30 solutions that were iterated 300 times for each population.
Distance threshold
dist
has been set to 50.0. Fuzzifier - µ values have been set to 2.5.
The other parameters have been given in Table 11.1. The algorithm flow as described in
Algorithm 3.
Population FFT - FFT-RECA The algorithm has been run 5 times, every time popula-
tion size was 30 solutions that were iterated 300 times for each population. Fuzzy threshold
value
f uzz
has been set to 0.15. The other parameters have been given in Table 11.2. The
algorithm procedure has been given in Algorithm 4.
Population FFD - FFD-RECA The algorithm has been run 5 times, every time popula-
tion size was 30 solutions that were iterated 300 times for each population. Fuzzy threshold
f uzz
has been set to 0.015. The parameters have been given in Table 11.2. The algorithm
implemented according to Algorithm 5.
In RECA algorithmic setting, as previously mentioned relevant parameters have been
introduced as described in Table 11.3. The presented parameters are distance threshold
dist
, fuzzy threshold
f uzz
, fuzzifier value µ, entropy α parameter.
11.5.3
Standard Indices - SI
In the experiments, the following standard indices have been considered and calculated for
the generated populations: Dunn index, DB index, KMEANS (kM), Otsu index, Turi index,
BETA index (β-index), within-class Variance - wVar, between-class Variance - cVar. The
description of the above algorithm has been presented in Section 11.3.
11.5.4
Parameters for RECA Measures
Rough entropy parameters that are applied in described in Section 11.4 algorithms depend
on the selected algorithm type and give possibility to adjust parameters to the concrete seg-
mentation task that most often will be influenced by imagery type, resolution, dimensional
context and other features. In order to make the present research and algorithm evaluation
the most informative each rough entropy algorithm has been assigned relevant parameters
that varied - distance threshold parameter has been valid only for CCD and FCD RECA
algorithms, fuzzy threshold has been valid for FFT and FFD RECA algorithms.
The rough entropy parameters have been considered in case of CCD RECA and FCD
RECA as described in Table 11.1. Each of these two algorithms has been performed in with
five independent sets of parameters. In the paper, these parameters sets are referred to as
R1, . . . , R5 set for CCD RECA algorithm and as F 1, . . . , F 5 for FCD RECA algorithm. In
case of FFT and FFD RECA algorithms the relevant parameters have been presented in
Table 11.2. In the same way as in previous two algorithms, parameters sets for FFT RECA
Search WWH ::
Custom Search