Particle area calculation

Is there any way to analyze complex images, as given(1), to determine area of each sepatate particle?
1.tif (2.7 MB)
Have tried method that i have used before, with thresholding-binary-fill holes-watershed, but this only works for ‘simple’/not packed imges, as given(2-3).
2.tif (1.6 MB) 3.tif (2.0 MB)
‘Fill holes’ for packed image gives me following(4-5)
4.tif (2.0 MB) 5.tif (2.0 MB)
So i guess i need to separate each particle by hand, which is very tidious and exhausting, and i`m wondering, is there any way to do this by the means of programm?

Have you tried the Weka trainable segmentation? Though after you generate particles, you might need to remove some based on circularity. It definitely looks difficult to tell what is “inside” versus “outside” in the image, especially given the overlap between objects.

Is there any chance you could take the images against a different background (and I don’t know what these are :slight_smile: )? If the objects were white but the background were… well, anything else, it would make the task a lot easier, I think.

Thank you for your reply
Yes, i have tried weka, but for some reason, its giving me a hard time to load made classifier. Like its loading for couple of hours
I`m using manual microscope that requires direct light, so i cant change the background
P.s. just to clarify - those are starch granules and weka gives me this result https://imgur.com/a/jrjowT3

Even if it is a manual microscope, it might be possible to put some kind of thin filter on the stage(* depending on where the camera is relative to the stage) Otherwise, hopefully someone else has some better suggestions. The requirement for context makes me think of deep learning, but I suspect there are other ways. It looks very similar to problems with phase contrast or DIC cell segmentation, so it might be worth looking into posts or methods on segmenting those.

Also slightly far afield if you already have your images, but I saw this post on researchgate, which led me to this as another potential option (or any other method of staining), if you are planning to do this across a large quantity of images. Fluorescent object detection is pretty much always easier, if it is an option.

@retgar

but did TWS give you the results you wanted otherwise? ie - you were able to build a classifier - but then just not apply it efficiently to other images?

There is a great BeanShell script that will batch process for Trainable Weka - applying a classifier to all images in a folder.

Thank you for your time
Unfortunately, TWS didnt make it for me so far. My machine simply cant handle big classifiers which i have to build for the reason that contrast of images can be different, due to the change of natural light. So i have to build classifier based on a big chunk of data i.e. many different photos

Sad to say but i already got a chunk of images made. and still, despite this method is good, i cant use it
but thank you for your help