ImageJ Segmentation when particles are too close together

Hello all,
I’m working on an independent experiment, and I’m attempting to separately segment the granules out of an SEM image of a material. The issue is that the granules are too close together and the lighting is non-uniform.

I’ve tried the Threshold tool, that doesn’t work well since the outlines of the granules are a wide swath of dark shades rather than just one shade, and the granules are too close together.

I’ve tried a tool called Weka Trainable Segmentation, which allowed me to trace the granules and outlines manually, and create a binary image that seemed to work, however not only was this method tedious with long wait times, but it also wasn’t precise enough as I repeated the measurements a handful of times, and the mean area distribution of the granules changed significantly each time after analysis.

The closest I got was adjusting the brightness of the image using “Curves” in photoshop, which raises the dark portions of the image up and flattens the lighting so all granules look like they’re in a uniform light, and THEN I used a combo of “Find Maxima” and “Threshold” on the darkest areas, and used Image Calculator with an operator of AND to combine both images. That gave the closest approximation of outlining the particles, but it still wasn’t 100% perfect.

So, as a plan B, I was recommended to seek help from an imageJ help website, and now I’m here. I would post a picture of the image I’d like to segment to show you guys what I’m trying to do, but I guess I can’t attach images because I’m a new user?

Hello Jacob -

A sample image would indeed be very helpful. Would it be possible
for you to upload the image to a file sharing site and post the link
here? (It’s best if you can avoid using a lossy format – such as
jpeg – for any sample images you share here.)

(As an aside, I probably won’t have any good ideas for your
segmentation problem, but I expect you will get a number of
useful suggestions from other forum participants.)

Thanks, mm

Hi, Here’s a link to 2 example images that I’m attempting to segment.

Hello Jacob -

I’ve taken the liberty of posting your two original images directly on
the forum:

StartingImage_largergranules.tif (949.2 KB)

Inverse FFT of Sample2_image_010.tif (949.2 KB)

And, since .tif’s don’t always display nicely on the forum, here are
the two images converted to .jpg’s:

Thanks, mm