DT-Binarize - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 156

(3)
(4)
(5)
(6)
4.3 Results
In the experimentation stage we generate four binary images using the three binarization tech-
niques (g90, g100, and Otsu) and DT-Binarize (see Figure 8 ) . Correctness of each binary im-
age is measured based on reference binary images (ground truth). Four different correctness
measures are employed at this stage in order to evaluate the results objectively. This process is
done for all test images in the dataset. Table 1 shows the average results of each measure. DT-
Binarize achieves 98.44% accuracy on our protein dataset, but accuracy is not a reliable meas-
ure for our dataset since true negative pixels are much more than true positive pixels in the
image. Therefore, F-measure, MCC, and Jaccard similarity measures are more reliable to eval-
uate DT-Binarize. According to Table 1 , our method could reach 81.06%, 82.36%, and 71.03%
for F-measure, MCC, and Jaccard respectively. DT-Binarize outperforms all other methods by
10% on the average.