Page 28 - IJEEE-2022-Vol18-ISSUE-1
P. 28
24 | Atiyah & Ali
In canny edge detection, noise reduction with a gaussian segmentation purposes, this technique looks for small or
filter is extremely important. It uses a gaussian filter, as these large pieces in an input image. We are using the “watershed
noises can be seen as borders, to remove noise from the transformation” to try a region-based method. Firstly, a map
image. After all, the edge detector changes suddenly in with the image's "Sobel gradient" can be found. The
intensity. When the image is flattened, the Ix and Iy background and image markers are also available based on
derivatives are measured concerning the ??-axis and y-axis. It the extreme part of the histogram of the gray value. Finally,
can be carried out with the image convolution of the Sobel- to fill the elevation map regions starting with the above
Feldman kernel. The dimensions of the gradient and the markers we use a “watershed transformation”. This last
angle of this kernel continue the process. Non-maximum procedure works better and the images can be individually
removal of falsified edges aim is to reduce the duplicate segmented and labeled. Figure 5 shows the image after
pixels fused around the edges so they are uniform. If the region detection.
existing pixel is greater than the magnitude of its neighbors,
the magnitude of the pixel intensity is set to zero. Two
specified threshold values are compared to gradient sizes, the
first is below the second. The gradients below the lower
threshold are removed, the gradients above the high
threshold high, and the pixels on the last border map are
shown. All the other gradients are marked as weak gradients
and pixels corresponding to them are taken into account in
the following steps.
Due to its connection to a strong edge pixel of a weak
edge pixel induced by real edges, the weak pixel will be
marked as an edge and is only used in the final edge map if
it is the same as some pixel with a high gradient. We will
develop and implement an algorithm that only once takes
account of all the coordinates of the gradient map. The final
edge map then allows deciding which pixels are included.
The final edge map then allows deciding which pixels are
included. Figure 4 shows the image after edge detection.
Fig. 4: Image After Applying Edge Detection Fig. 5: Image After Applying Region Detection
E. Region-based Segmentation F. Training and Implementation
This approach involves dividing an image into regions
The BraTS dataset includes MRI images and the
equivalent to a variety of constraints. The region-based associated segmentation results for training and model
segmentation technique includes an algorithm that divides an testing. Neuroradiologists who are clinically qualified
image into different components with related pixels. For correct images marked output or ground truth. The train data
and validation data folders are in the BraTS dataset. Each
subfolder contains 5 sophisticated pictures of five different
models, such as T1, T1ce, T2, T2-FLAIR, and Seg. The data
folder contains 369 subfolders. In the training dataset, a total
of 1845 images. A data validation folder contains 125
subfolders and four images in four different modes such as
T1, T1ce, T2, and T2-FLAIR in each subfolder. In the
validation data folder, there are a total of 500 images.
After evaluating the data, the images are pre-processed
by standardizing the intensity value and cropping them.
Applied edge-based or region-based algorithms before
augmenting the dataset. MRI records are split into three
categories: train set, test set, and validation set. 60% of the
images are used as the train set, 20% are used as the test set,
and 20% are used as the validation set. The training images
are data augmented, which aids generalization and enhances
accuracy. The albumentation library uses techniques of
enhancement, such as grid distortion, random brightness
