Is there a way to make a stack of images look the same (brightness/contrast/color) to apply WEKA segmentation for better results?


I am using a lot of microscopy images of fish fins and using trainable weka segmentation in order to classify different colored pigmentation cells.
However, I have to keep adjusting when I upload data for each new image because the images have different brightness/contrast/overall color tones.
Is there a way to correct for this and get all the images to look relatively similar to each other so that the trainable weka can identify what I want better?

Thank you so much in advance!

It depends on what is in the images. There is some histogram normalization, but that can fail horribly if the images are actually different. ImageJ has some options for white balancing, but I think they also rely on having a true white object in the image to use as a reference.

The best option, of course, is to fix things at the data acquisition end. =/ Images will never really be as comparable after processing, and automated methods of calculating “stuff” are more sensitive to problems with image acquisition than we are. So the input quality needs to be greater.

Hello seulbs,
You could try making a copy of your best image in the stack. then goto Process > Image Calculator and put your stack as Image 1 and your copy as Image 2 and set the function to average. This can be repeated several times if necessary.