DT-Binarize - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 152

FIGURE 3 Contrast stretching example: (a) original image and (b) image after applying con-
trast stretching.
3.2.3 Decision tree
Decision tree [ 10 ] is a rule-based classifier that employs a tree structure for data classification
It is a supervised classification technique that comprises of training and testing stages. In the
training stage the tree is generated based on the entropy of the data features. In the testing
stage, each test sample is classified using the tree built in the training stage. Decision tree is a
classifier that requires relatively less time to create training model. In addition, testing is quite
fast after building the tree.
3.3 Application of Dt-Binarize on Protein Crystal Images
Protein image binarization problem is a convenient application area of DT-Binarize, since
there is no single thresholding method that can generate proper binary images for all datasets.
In our problem, we labeled the protein crystallization images with one of the three diferent
thresholding methods: Otsu's threshold 90th Percentile Green Intensity threshold, and Max
Green Intensity threshold. We use the training images to build the decision tree based on only
standard deviation of the pixel intensities. 75% of the data is selected as the training set, and
the remaining is used for the testing. Figure 4 shows the decision tree of the training stage.
In Figure 4 , “g90” is selected as the best binarization method if standard deviation of the test
sample is less than 12.86. However, if the standard deviation is between 12.86 and 24.99, the
best binarization method is selected as “Contrast Stretching + g90.” Similarly, other binariza-
tion methods may be selected depending on the standard deviation of the test image.
FIGURE 4 Decision tree for selecting the best threshold method.
We have employed this tree to our test dataset. For a test sample, we take the standard de-
viation and find the corresponding class label of the tree. The method represented with that
label is selected to binarize the test image. The following section provides some numerical and
visual results of DT-Binarize with several examples.