You are on page 1of 1

Region of Interest Detection in Fundus Images Using Deep Learning

and Blood Vessel Information


Step 1: Extract the red channel from the image and reduces noise in the red channel.
Step 2: Convert the red channel into a binary image using Otsu's thresholding. This binary image
distinguishes the retina area from the background.Apply the Canny edge detection operator to the
binary image to create an edge image.
Step 3: Utilize OpenCV's ellipse detection function to estimate an ellipse from the edge image. This
ellipse represents the approximate shape of the optic disc and helps determine the longest diameter
(d) of the ellipse.
Step 4:Use a median operator with a window size of [3x3] to reduce noise.
Apply CLAHE histogram equalization to enhance the image's contrast, resulting in a contrast-
enhanced image.
Step 5: Calculate a difference image by subtracting the contrast-enhanced image from another image.
This subtraction is subject to certain conditions: another image must be greater than contrast-
enhanced image, and binary image must be greater than 0; otherwise, set the pixel to 0.
Step 6: Apply morphological closing using a [3x3] square to the difference image . This operation fills
in any missing blood vessel pixels inside the vessels.
Normalize the result from the previous step to the range [0, 255], resulting in a normalized image .
Step 7: Apply Otsu's thresholding to the normalized image, creating a binary image. This step helps
remove pixels belonging to minor blood vessels and noise.
Remove smaller blob pieces in the binary image with fewer than 20 pixels.
Step 8: Generate a gray-level blood vessel image based on the binary blood vessel image. If a pixel in
binary blood vessel image is greater than 0, set the corresponding pixel in gray-level blood vessel
image to 255.
Step 9: Create an image by embedding the gray-level blood vessels image into the original fundus
image. Add the blood vessel pixels to the respective color channels of the original image if the green
channel value is greater than the corresponding pixel in the gray-level blood vessel image.
Step 10:After completing these steps, you have a final image that contains the blood vessels
embedded in the original fundus image.

You might also like