Brain Tumor Detection Using Image Processing Techniques

127

FIGURE 4.5

Adaptive thresholding techniques were applied on same brain MRI images.

The source image featured in this figure was selected from the dataset available

as open source on Kaggle [5].

In the next step, the most appropriate threshold value is determined.

Otsu’s method aims to identify the threshold that minimizes the weighted

within-class variance, essentially equivalent to maximizing the between-class

variance. The threshold value that meets this condition is the optimal thresh-

old value used in image segmentation. Otsu algorithm includes the following

steps [41]:

Algorithm 2 : Otsu Algorithm

1: Grayscale image is taken

2: The histogram of the grayscale image is calculated

3: The pixel density probabilities of the image are found

4: Initial values are assigned for wb(0), wf(0), µb(0), and µf(0).

5: The following steps are applied for all threshold values from t=0 to the

highest pixel intensity value

6: wb(t), wf(t), µb(t), and µf(t) values are updated

7: The between class variance σ²

B^{(t) is calculated}

8: The threshold value of t at which σ²

B^{(t) is maximum is determined}

Adaptive Thresholding Method

The adaptive thresholding method divides an image into smaller regions and

computes the threshold value for each region. Thus, every region will have

different threshold values. For each region, the threshold can be computed

either using arithmetic mean or Gaussian mean of the pixel intensities. In the

arithmetic mean, each pixel in the neighboring region contributes the same

amount to the threshold calculation. In the Gaussian mean, the pixel positions

play a significant part in the threshold calculation. Pixels that are further away

from the center of the region are less likely to contribute to the calculation.

Figure 4.5 demonstrates the resulting output images after applying adaptive

thresholding techniques.