Changing contrast and Segmentation

segmentation

#1

Hi all,
I’ve been trying to segment a component within my image but I think the contrast is not sufficient for the threshold method to work. An example of the image is here

you might see in the image some circular ish regions (which is not the gap in the middle, that gap is an artery with a vein besides) surrounded by bundles.
The circular regions are pores within the tissue surrounded by bundles of collagen. a schematic of the image is here
schematica

in this image you see the circular pores and the surrounding bundles.
I am trying to segment the circular regions from the rest of the image (from everything else).
From the first image I upload it’s very tricky, i tried to apply different filters but nothing, the threshold method cannot pick up the circular pores as they grey level is very similar to that of the surrounding everything else.
does anybody know a way of doing this?

I would like to obtain something like this

where all the circular regions are black and separated from everything else.
I spent lots of time on this but nothing good came up.
Any suggestions?
Thank you


#2

Good day,

please post the original raw image, i.e. without any preprocessing.

What do both areas in the posted sample image mean that are surrounded by a white line?

Regards

Herbie


#3

Hi,
that you for your quick response. This is the raw image (an example of slice within a bigger stack). It’s not the same uploaded before but does not matter, they are all similar.

Without filtering it, it’s even more complicated. it’s very tricky to identify the circular regions.

I think that the white line surrounding the two holes in the middle is nothing. I mean, it’s the inner layer of artery and vein so in reality it’s connective tissue. Why it comes up so bright? I don’t know.
Also in this one, there are gaps at the top as well which are vessels too. They can be cropped away.

Thank you


#4

Thanks for the image and the details.

I shall have a closer look at the image and the processing options in a few hours.

Regards

Herbie


#5

Many thanks.
Will wait for your suggestions!

Regards


#6

Sorry Laura,

but except for the irrelevant and high contrast details I feel unable to extract anything of relevance from the sample images.

I have no idea how they were acquired.
Some artifacts look a bit like strange CT-artifacts but I doubt that these are CT images. There are reasons to assume that they are EM-images …

If you have gathered sections (you speak of a stack of images) whose contents doesn’t chnage much from slice to slice, you may try to average over 4 slices which at least may reduce the noise by a factor of two (provided the noise is uncorrelated random) and may give better contrast.

In the first place however, I strongly suggest to improve the image acquisition process. (Ask colleagues who are experienced with the same way of image acquisition!)

Good luck

Herbie


#7

it’s ok, no worries, thanks a lot for having tried anyway.
These are XMT image (x ray microtomography), so I guess similar to CT. No chemical contrasts were added to the sample prior acquisition and it was a preliminary experiment to develop a method. Images did not come out very good.
Will try to average every 4 slices, perhaps I can get something out of it.

Many thanks, very much appreciated.

Regards


#8

Microtomography sounds plausible …

Concerning the artifacts:
You won’t get rid of them because they are caused by structures that give rise to broadband-signals that violate the sampling theorem.

Concerning the noise:
Use a higher X-ray dose (more X-ray photons) if this doesn’t destroy the structures.

Concerning the contrast:
Play with the energy of the X-ray source (“color” of photons).
(I guess lowering is the way to go.)

Have success

Herbie


#9

I am working on something similar. my focus is these small round items but its pretty difficult. any suggestions?


#10

Good day!

It has been told here on the Forum a 100 times that JPG-compressed images are unsuited for any kind of scientific image evaluation. This holds especially true for the kind of noisy images that you are interested in!

So please post a typical raw image in the original TIF- or PNG-format. No JPG-format though, because JPG introduces artifacts! You may also post images as Zip-archives.
(Converting a JPG-compressed image to TIFF- or PNG-format doesn’t make sense.)

Regards

Herbie


#11

Laura.biology,
Try this to see if the contrast is acceptable.
Bob


#12

My Apologies, I should read all previous messages regarding what type should be posted. Sorry!


#14

Thanks for the image!

After having spent about 90 minutes trying to get a reasonable result with standard methods, I must admit that this appears impossible.

As you may have found out yourself, the illumination of the structures in question is uneven in a sense that mainly the left-bottom border parts are brighter than the remaining border parts. This makes it really difficult to get closed contours. If you scan the Forum, you will find a lot of similar cases and the best remedy is to optimize the image acquisition process.

Good luck

Herbie


#15

Very Herbie,
Thanks for your time, then it means i need to create my own macro to resolve the issue. will keep researching on what method will make it. However, I would like to admit that, that was the image sent to me.
Thank You once again.
reuben


#16

[…] i need to create my own macro to resolve the issue.

I really wouldn’t bother trying to do so. I’m pretty sure you won’t come up with a decent result and I’m not sure if more advanced classification-based approaches will provide such. That said, you may try WEKA which comes with FIJI.

I would like to admit that, that was the image sent to me.

This means, you didn’t acquire the images yourself hence, you are unable to change the acqusition process?

If so, contact the person who did the investigations and tell her/him what makes the image analysis difficult.

Regards

Herbie


#17

thank you.
will keep you posted.
My Prof gave the image to me to do a threshold and segment and later calculate the area and do other analysis on the circular items in the image
I failed to mention that it is SEM of a metal, but proposed i try using FIJI .
ONCE again thank you.


#18

I was thinking of writing a macro because when you go to PROCESS-FIND MAXIMA , you will realise that it selects the particles.

Check out the screenshot i attached, I was thinking if this is the case, there should be a way to classify just that at once without taking it one particle at a time.


#19

Sure, you may find parts of the objects of interest but you like to get them in a way that allows for size measurements etc.

Here is an example of a first attempt of contour tracing:
K_WechatIMG718-2
However, I wouldn’t call this approach a standard method …

[…] there should be a way to classify just that at once without taking it one particle at a time.

No idea what you mean.

Regards

Herbie


#20

smithrobertj

it is better of course but probably not enough to use the treshold method to segment.
Thank you


#21

what i meant about the classification was thresholding. still reading on your suggestion about the WEKA. If i can get a good thresholding then i can do a perfect segmentation and do any analysis possible.

Thanks Anyway.
Means a lot.