Image Segmentation: A Rough-set Theoretic Approach - Rough Fuzzy Image Analysis: Foundations and Methodologies - page 194

Since the peaks in the roughness index based method occur exactly at the in-
tensities where the number of similar intensity pixels is large as compared to the
number of same intensity pixels, the color of the every segmented region is more
close to the color of the corresponding region in the original image.
10.4SegmentationMethod
The segmentation process is divided into three stages, as shown in Figure 10.2. For segmen-
tation we have considered RGB color space. In the rst stage, the histogram and the histon
of the R, G, and B components of the image are computed. For computing the histon,
the selection of two parameters is very important. These two parameters are: the neigh-
borhood and the expanse. Neighborhood is the window that decides the pixels involved in
the computation. For example if a 3 3 neighborhood is selected then the pixels used in
computation of histon will be 3 3 1 = 26. The expanse is the radius of the similar color
sphere. After sucient experimentation we found that selecting a 3 3 neighborhood and
an expanse value of 100 gives best results. The roughness index is then obtained for R, G,
and B components of the image, for every intensity value, using equation (10.11).
FIGURE 10.2: Block diagram of the proposed method
10.4.1SelectionofpeakandThresholdvalues
Like histogram, the roughness index for all the intensities also gives the global distribution
of uniform regions in the image and every peak represents a uniform region. In the sec-
ond stage, we determine the peaks and valleys in the graph of roughness index and apply
thresholds to the image. Selection of correct peaks and valley points are very important for
achieving good segmentation results. The criterion used for selection of signicant peaks is
based on distance between two signicant peaks and the height of the peak. Experimentally
we have found the following two criteria for obtaining the signicant peaks:
1. The peak is signicant if the height of the peak is greater than 20% of the average
value of roughness index for all the pixel intensities.
2. The peak is signicant if the distance between two peaks greater than 10.