Image Processing Reference
FIGURE 9 Example of a bad binarization result: (a) original image, (b) ground truth image,
(c) g90 threshold, (d) g100 threshold,(e) Otsu threshold, and (f) DT-Binarize.
Please note that success of DT-Binarize depends on the thresholding methods that are
chosen for the problem domain. If none of the thresholding techniques provide a correct res-
ult, our method will not provide a good result either. It means that there is a performance
limit that DT-Binarize can achieve. We call this limit as the “Max Limit.” If we select the
best thresholding method for every image out of the provided techniques, we can reach the
“Max Limit.” In this case, DT-Binarize is considered to perform well when it chooses the same
thresholding techniques that are chosen manually. Figure 10 shows the comparison of the cor-
rectness of our technique with respect to the best labeling. The closeness to the max limit in-
dicates the success of DT-Binarize in this problem. Numerical results are presented in Table
2 . The max limit of F-measure is 85.2%, and DT-Binarize could achieve 81.06%. It means that
DT-Binarize can reach 95% of the max limit for protein image dataset.
FIGURE 10 Comparison with theoretical max limit.