Image Processing Reference
In-Depth Information
FIGURE 10.4: Segmentation comparison. (a)Test image92059(b)Histogram based seg-
mentation (c)Histon based segmentation (d)Roughness index based segmentation
S k . The Probabilistic Rand Index (PRI) is dened as:
N 2
PRI (S t ;fS k g) =
[c ij p ij + (1 c ij ) (1 p ij )]
i6 = j
where c ij denotes the event of a pair of pixels i and j having same label in test image
S t : c ij = II
l S t i = l S t
and p ij denotes the ground truth probability that II (l i = l j ):
p ij = 1 K P II
l S i = l S t
. This measure takes values in [0; 1], where 0 means S t and
fS 1 ;S 2 ;:::;S K g have no similarities and 1 means all segmentations are identical.
One can make several observations from the visual and the quantitative comparisons.
The colors of the segmented regions look more natural in case of roughness index based
thresholding algorithm. In case of the test image22013of Figure 10.3, the color of sky
Search WWH ::

Custom Search