Image Processing Reference
FIGURE 2 Overview of the DT-Binarize.
1. resizing and noise reduction of the images,
2. labeling training images with best binarization methods,
3. extracting statistical features of the images,
4. building the decision tree based on the statistical features,
5. predicting the best binarization method for a test image using the decision tree,
6. applying chosen binarization method to a given test image.
3.2 Stages of the Algorithm
3.2.1 Median filter
Median filter is one of the well-known order-statistic filters due to its good performance for
some specific noise types such as “Gaussian,” “random,” and “salt and pepper” noises [ 3 ]. Ac-
cording to the median filter, the center pixel of a M × M neighborhood is replaced by the medi-
an value of the corresponding window. Note that noise pixels are considered to be very difer-
ent from the median. Using this idea median filter can remove this type of noise problems [ 3 ].
We use this filter to remove the noise pixels on the protein crystal images before binarization
3.2.2 Contrast stretching
Contrast stretching is a normalization method that enhances the informative features of the
image by expanding the histogram of the intensities. It maps the pixel values into a new range
where I in and I out are the input and output images, P min and P max are the minimum and the
maximum intensity value of the input images, and P′ min and P′ max are the minimum and the
maximum intensity values of the output image, respectively. We include contrast stretching
in our research, because our dataset contains some low contrast images that may cause incor-
rect thresholding for our dataset. Figure 3 shows a problematic image and contrast stretching
result. Note that informative features of the result image are magnified without loosing the
structure of the crystal, and contrast stretching is not applied all the images in the dataset.