Image Processing Reference
FIGURE 9.4: GUI of the 'perceptual images' system
data base panel at the left side offers these user friendly options to the users. Feature se-
lection is also a part of this panel. Figure 9.5 shows an example of two different images and
their coverings. The other two panels at the bottom are responsible to choose the analysis
methods. In the analysis panel user can chose between different image comparison methods
such as tNM, TOD, and TCD. The results will be showed in the results panel.
Figure 9.6 shows another example of the images created by the GUI. First, the original
images will be loaded to the GUI. Then, by choosing gray scale as a feature, the image
will be converted to gray scales, and then the tolerance classes (covering) will be produced.
User can choose the windowing size and epsilon as well as which measures to be calculated.
Tables 9.2, 9.3, and 9.4 show the results of comparison of different images in Figures 9.1,
9.2, 9.5, and 9.6 with three different measures. The set of images include 8 different images
which are numbered from 1 to 8 for simplicity. Each column of the tables show the results
of pairwise comparison of one specific image to the rest of images. Table 9.2 demonstrates
the results of TOD nearness measure, while Table 9.3 presents the results for tNM nearness
measure. Finally, the results of TCD nearness measure is included in Table 9.4.
Figures 9.7, 9.8 and 9.9 show three example plots for the comparison of images using
three nearness measures TOD, tNM, and TCD. In each plot x-axis shows images using
their numbers, and y-axis shows the value of different measures. Figure 9.7 shows the re-
sults of comparison of image 1 (Lena) to all the other images. As it can be seen on the plot,
on number 3 for example, all the measures indicate that images 1 and 3 are very different,
or the similarity of image 1 with image2 (number 2 of x-axis) is very high, which is very
similar to what is expected. On the other hand, for images 5 and 6, the results of tNM
is significantly different from two other measures TOD and TCD. tNM shows that images
5 and 6 are quite different from image 1, which is closer to the expected results. Hence,
tNM shows a more accurate result in this case. Figure 9.8 shows another example that
tNM works better than TOD and TCD. However, since TOD and TCD's results depend on