Image Processing Reference
In-Depth Information
(10.5a)
(10.5b)
(10.5c)
(10.5d)
FIGURE 10.5: Segmentation comparison. (a)Test image97017(b)Histogram based seg-
mentation (c)Histon based segmentation (d)Roughness index based segmentation
and water regions in the roughness index based segmentation, see Figure 10.3dd, look more
natural in comparison of those regions in the histogram based and histon based segmenta-
tions of Figures 10.3bb and 10.3cc. In case of test image92059of Figure 10.4, the regions
like river and boat are well segmented in roughness index based segmentation as compared
to histogram based and histon based segmentations. Consider a test image271031of Fig-
ure 10.8. It can be observed that the Sun is clearly and neatly segmented by the proposed
method, see Figure 10.8dd, and not so neatly segmented by the histogram based and the
histon based segmentations of Figures 10.8bb and 10.8cc.
The quantitative comparison of the these images is given in Table 10.1. It can be observed,
from the PSNR and PRI values, that roughness index based segmentation results are better
than the other two segmentation methods. The number of clusters after merging is nearly
the same as in case of histogram based segmentation.
Computational eciency of the algorithm depends on several factors such as computa-
tional complexity histon, obtaining the threshold values and region merging process. The
major computational cost is involved in region merging process. The computation time
is directly proportional to the ratio of the number of clusters before and after merging.
The Table 10.1 also shows computation time of all the algorithms. It can be observed
that whenever the ratio of number of clusters before merging and after merging is high the
computation time is more. This means that the additional burden involved in computation
of histon does not add to the computational cost of the complete segmentation process
signicantly.
10.6Summary
 
 
Search WWH ::




Custom Search