Image Processing Reference
In-Depth Information
TABLE 12.5
Similarity Measurements
Metric Feature
Images
ε
tNM-measure Hausdorff-measure
L
2
greyscale
Im1, Im2
0.5
0.49646
0.78257
L
2
greyscale
Im1, Im3
0.5
0.58015
0.87151
L
edge-o.
Im1, Im2
0.5
0.60141
0.88289
2
L
edge-o.
Im1, Im3
0.5
0.63080
0.88147
2
L
greyscale
Im1, Im2
0.5
0.49646
0.78257
1
L
1
greyscale
Im1, Im3
0.5
0.58015
0.87151
L
1
edge-o.
Im1, Im2
0.5
0.60141
0.88289
L
1
edge-o.
Im1, Im3
0.5
0.63080
0.88147
L
1
greyscale
Im3, Im4
0.5
0.74525
0.92248
L
edge-o.
Im3, Im4
0.5
0.88960
0.92419
1
L
edge-o.
Im3, Im7 0.5
0.91027
0.92269
1
L
greyscale, edge-o. Im1, Im2
0.5
0.44694
0.78119
1
L
1
greyscale, edge-o. Im1, Im3
0.5
0.52776
0.77003
L
2
greyscale, edge-o. Im1, Im2
0.5
0.44644
0.76412
L
2
greyscale, edge-o. Im1, Im3
0.5
0.52776
0.75469
L
1
greyscale, edge-o. Im3, Im4
0.5
0.67706
0.80572
L
greyscale, edge-o. Im3, Im5
0.5
0.94620
0.76481
1
L
greyscale, edge-o. Im3, Im6
0.5
0.92052
0.80602
1
L
greyscale, edge-o. Im3, Im7
0.5
0.89572
0.80945
1
L
1
greyscale, edge-o. Im3, Im8
0.5
0.90605
0.81532
L
1
greyscale, edge-o. Im3, Im9
0.5
0.90262
0.79911
L
1
greyscale, edge-o. Im3, Im10
0.5
0.86576
0.78755
L
1
greyscale, edge-o. Im3, Im11
0.5
0.92289
0.79546
L
greyscale, edge-o. Im3, Im12
0.5
0.77348
0.75115
1
L
greyscale, edge-o. Im3, Im13
0.5
0.83372
0.78273
1
L
greyscale, edge-o. Im3, Im14
0.5
0.82593
0.78242
1
L
1
greyscale, edge-o. Im3, Im15
0.5
0.96127
0.81383
L
1
greyscale, edge-o. Im3, Im16
0.5
0.79675
0.74366
L
1
greyscale, edge-o. Im3, Im17
0.5
0.81891
0.76443
L
1
greyscale, edge-o. Im3, Im18
0.5
0.85555
0.78585
L
greyscale, edge-o. Im3, Im19
0.5
0.92766
0.79726
1
L
greyscale, edge-o. Im3, Im20
0.5
0.83858
0.80834
1
image Im4= 0.677.
Figure 12.6 shows the histogram values for closest and farthest image for the two features. It
can be observed that the greyscale feature values fall into two ranges and the edge-orientation
feature values fall into three ranges. Their combined influence is what is observed in the
overall tNM measure. The Thesholded Images areshown in Figure 12.6b, Figure 12.6f and
Figure 12.6j.
12.5.2
Hausdorff-measure Results
In this section, the experiment with sample of 17 images compared with the query image is
repeated in Figure 12.7a using the Hausdorff measure. The sub figures that form a part of
Figure 12.7a are arranged in descending order of their Hausdorff measure values with the
nearest image Im8 = 0.815 and the furthest image Im16= 0.744.
It can be observed that Hausdorff measure values are fairly consistent and show little
variation between the nearest and the farthest image. Figure 12.6 shows the histogram
values for closest and farthest image for the two features. The two sets of experiments also
demonstrate completely different image retrieval results for the two measures based on two
features.
It can also be observed from Table 12.5 that the two measures based on edge-orientation
and L
metric would have almost similar values for Im7 for query image Im3.
1
Search WWH ::
Custom Search