Image Processing Reference

In-Depth Information

Analysis of Performance

Consider Figure 3.12 that shows box plots (Tukey, 1977) which graphically depict the LCE

values corresponding to the segmentation achieved by bilevel thresholding using algorithms

(i) to (vii) mentioned earlier in this Section. A box plot, which in Figure 3.12 summarizes

the LCE values obtained corresponding to all the segmentation ground truths available for

an image, is given for all the 100 images considered. We find from the box plots that the

LCE values corresponding to the algorithms (i), (ii), (v) and (vii) are in general smaller

compared to that corresponding to the other algorithms. It is also evident that algorithms

(ii) and (vii) perform almost equally well, with algorithm (ii) doing slightly better.

From all the box plots in Figure 3.12, we find that segmentation results achieved by

algorithms (i) and (v) are equally good and they give the best performance among the

algorithms considered. The average of all the LCE values obtained using an algorithm is

minimum when algorithm (v) is used. However, maximum number of zero LCE values

are obtained when algorithm (i) is used. Hence, we say from quantitative analysis that

algorithms (i) and (v) are equally good and they give the best segmentation results.

3.6

Conclusion

In this chapter, image thresholding operations using rough set theory and its certain general-

izations have been introduced. Classes of entropy measures based on generalized rough sets

have been proposed and their properties have been discussed. A novel image thresholding

methodology based on grayness ambiguity in images has then been presented. For bilevel

thresholding, every element of the graylevel histogram of an image has been associated with

one of the two regions by comparing the corresponding errors of association. The errors of

association have been based on the grayness ambiguity measures of the underlying regions

and the grayness ambiguity measures have been calculated using the proposed entropy mea-

sures. Multilevel thresholding has been carried out using the proposed bilevel thresholding

method in a binary tree structured algorithm. Segmentation and edge extraction have been

performed using the proposed image thresholding methodology. Qualitative and quantita-

tive experimental results have been given to demonstrate the utility of the proposed entropy

measures and the effectiveness of the proposed image thresholding methodology.

Search WWH ::

Custom Search