Image Processing Reference
In-Depth Information
are described as T 1, . . . , T 5 parameters sets and FFD RECA parameters are denoted as
D1, . . . , D5. Additionally, rough entropy values calculated on the base of above-mentioned
rough entropy parameters referred to as rough entropy measures determine rough indices
RI that values are optimized by evolutionary algorithm in the search for optimal clustering
segmentation solutions as described in Algorithm 1. On the other hand, rough indices are
compared with standard clustering validity measures giving in-depth notion about internal
rough entropy properties.
TABLE 11.1 CCD RECA and FCD RECA measures - distance threshold
dist
, fuzzy threshold
f uzz
and fuzzifier -
µ
CCD
FCD
P
dist
P
dist
µ
R1 10 F1 50
1.5
R2 20 F2 50
2.0
R3 30 F3 50
2.5
R4 40 F4 50
3.0
R5 50 F4 50
3.5
11.6
Experimental results
11.6.1
The correlation between standard and rough measures
Correlation values have been calculated between standard validation indices SI and rough
indices RI independently for all four populations: CCD RECA population, FCD RECA
population, FFT threshold RECA and FFD RECA population as described in Subsection
11.5.2. In the comparison, in the first part, the satellite build-up area image has been
analyzed, in the second part, the nature image 78004 in low and medium dimensions has
been presented, and finally, the results for the 27059 nature image in high dimensional
context with 9 bands in low, medium and high resolutions are gathered.
Satellite buildup area
Satellite buildup area image have been considered in low dimensional setting: bands RG,
RB and BG and in the medium dimensional setting: RGB band. The indices correlations
for four selected populations are presented in Table 11.4 and Table 11.5.
Results for the selected low dimensional setting and medium dimensional context are
presented in Table 11.4 in the context of standard validity indices correlation and in Table
11.5 in the rough indices correlation. In the class of standard validity indices, Dunn index
has been maximally correlated with Turi index, Davies-Bouldin index has achieved the
best correlation with β-index, wVar and cVar indices.
The β-index has been most often
correlated with wVar and cVar indices.
TABLE 11.2 FFT RECA and FFD RECA measures - distance threshold
dist
, fuzzy
threshold
f uzz
and fuzzifier -
µ
FFT
FFD
P
f uzz
µ
P
f uzz
µ
T1 0.3
1.5 D1 0.3
2.5
T2 0.3
2.0 D2 0.03
2.5
T3 0.3
2.5 D3 0.01
2.5
T4 0.3
3.0 D4 0.001 2.5
T4 0.3
3.5 D5 0.0001 2.5
Search WWH ::
Custom Search