1

I am trying to classify infected red blood cells(RBC) and uninfected ones and am trying to do some image preprocessing that might help boost accuracy scores. I using this preprocessing for XGBOOST and SVM.

Asking for help here as my capstone tutor is not responding for quite some time.

Image segmentation example

I give two examples 1st on the left and 2nd on the right. My goal is to segment the infected places inside the RBC, the darker spots.

What I have currently done is:

  1. normalize the image
  2. get the histogram of the colored the normalized image
  3. if there is one peak of the channel then pick the last lightest value of the "hill" base. If there are two peaks on the channel then pick the value in-between the "hills".
  4. with the picked values of each channel segment the image in range from (1,1,1) to (red value, green value, blue value)

All of the steps above were done manually and they work (shown in the image link I gave).

I want to do this automatically as I have a huge data set.

My Questions:

  1. How do I get the base values where the peak ends automatically

Also I am using python.

EDIT: Sorry, I did not realize to add the images I'm working with. Here is the data set zip folder I am working with: https://data.lhncbc.nlm.nih.gov/public/Malaria/cell_images.zip

Here are separate images just in case:

infected_img

uninfected_img

infected2_img

infected3_img

uninfected2_img

Saulius
  • 11
  • 2
  • You might have more success with answers if you shared separate, actual input images that folk can test their strategies on, rather than low quality diagrams of images that people have to mess around extracting. – Mark Setchell Feb 08 '22 at 08:51
  • @MarkSetchell Thanks for the comment. I did not think about this, my apologies. I edited the post. – Saulius Feb 09 '22 at 13:07

0 Answers0