Image Processing Reference
In-Depth Information
TABLE 11.8 Standard indices correlations in RECA populations
for generic 27059 image - 9 bands
Name CCDRECA FCDRECA FFTRECA FFDRECA
9-High
Dunn DB-0.59 DB-0.65 DB-0.58 DB-0.61
DB Dunn-0.59 Dunn-0.65 Dunn-0.58 Dunn-0.64
β
-index kM-0.97 kM-0.95 kM-0.96 kM-0.95
9-Medium
Dunn DB-0.60 DB-0.65 DB-0.59 DB-0.65
DB Dunn-0.60 Dunn-0.65 Dunn-0.59 Dunn-0.65
β
-index kM-0.96 kM-0.95 kM-0.96 kM-0.95
9-Low
Dunn DB-0.60 DB-0.65 DB-0.76 DB-0.66
DB Dunn-0.63 Dunn-0.65 Dunn-0.76 Dunn-0.66
β
-index kM-0.96 kM-0.95 kM-0.96 kM-0.95
In the class of rough entropy validity indices, the Dunn index has been maximally cor-
related maximally with various RECA measures. In case of the Davies-Bouldin index pre-
dominantly the highest (negative) correlation has been observed for FCD and FFT RECA
measures. The best correlations for the β-index are achieved for the CCD and FCD RECA
measures. The comparison shows that β-index correlations are the highest from three se-
lected indices and the correlations for DB are are comparatively the lowest.
TABLE 11.9 Rough indices correlations in RECA populations -
generic 27059 image - 9 bands
Name CCDRECA FCDRECA FFTRECA FFDRECA
9-High
Dunn R50.10 T50.00 F50.08 T50.02
DB F5-0.04 T50.00 T4-0.05 T5-0.09
β
-index F20.67 R40.72 R50.69 R40.72
9-Medium
Dunn R50.12 F5-0.01 R50.11 T50.04
DB T5-0.02 T50.00 F4-0.14 T5-0.11
β-index F20.69 R50.71 R40.72 R40.74
9-Low
Dunn R50.12 R40.02 F40.28 T50.01
DB T5-0.04 T50.00 T4-0.30 T5-0.10
β-index R40.71 R40.74 F30.78 R40.71
Standard and rough index correlation - summary
The comparison performed for three datasets in low, medium and high dimensions sug-
gests that rough indices have tendency to be highly correlated with the β-index values.
This correlation correspondence seems to be the most profound in all data dimensions, but
exhibits slow decrease as data dimensionality increases. In case of Dunn and DB indices, the
moderate correlation have been observed for low dimensional datasets that tend to decrease
for medium and high dimensionality.
In the subsequent tables, the best solutions from rough populations CCD RECA, FCD
RECA, FFT RECA and FFD RECA are presented. The best solutions are considered
relative to Dunn index, Davies-Bouldin index and β-index. In table 11.10 the best standard
indices values for satellite buildup area image have been presented. Satellite buildup area
image ave been segmented in low dimensional context (bands RG, RB, BG) and medium
dimensional context (bands RGB).
Search WWH ::
Custom Search