Image segmentaion for biological images with low contrast

Dear All,

could any one please help me to give a technique or method to extract the biological objects from the image shown below

[Walter Roberson]

You are going to have to point out to us the boundaries of some of the biological objects so we can see what needs to be paid attention to and what is background.

[mayar aly]

The objects surronded by black line are an interested objects

[Walter Roberson]

When I increase the contrast, I find three clear dark circles with thin circle outlines, all in the upper right quadrant. I also see a number of fainter areas that might be circles. Or they might just be texture of the background.

Keep in mind that we don't know what we are looking at, so we do not know how to interpret what we are seeing.

In the image you attached, how many definite biological objects are there, and how many locations that "hint" sufficiently that you would be running some kind of test to determine whether they were significant or not?

Is the task to find as many circle-like features as possible ? Is a range of radii given? For example once we have picked out the most obvious of the circles, can we use its inner diameter as the radius to look for to find the other objects?

Do you have a sample input image and a sample labeling of the same image?

[Walter Roberson]

Your image had not appeared when I previously replied.

I have looked at the image with the circled objects, and I see absolutely nothing that would distinguish those from anything else on the image. I have no clue what you looked for. If you had picked any other random blob on the image and said that it was what you were looking for, I would not have been able to tell the difference. All I see are three clear dark circles in the upper right (after I adjust the contrast).

[mayar aly]

i will tell you what i had do

i used "Adaptive Histogram Equalization" and get the result as shown below

but, i don't know what is the best filter required to enhance this image ??

Also, what is the segmentation method is aplicaple to those images?

Thanks in advance..

[Image Analyst]

AHE is an enhancement. Why do you need further enhancement? What about denoising or segmentation instead? We still don't know what is in the three regions you outlined so please explain that.

[mayar aly]

i need further enhancement as denoising techniqus to pick out the noises.

i meant by these regions, the interested objects.

at last image, the interested objects are the circular and dark objects

[Image Analyst]

Is this image the same as the one a few comments ago where you outlined the reqions? I don't see any dark circles in that image inside the regions you outlined.

[mayar aly]

Ok, you are right .. the first image has an interruption while i upload it..

The original image is

Then, the adaptive one is

Then, i try to use the average filter

Then, adjust contrast

but, when i try to use an active contour as segmentation method, the result is never accurate

Answer 1

[Image Analyst]

Your best method I think is to freehand trace them with imfreehand(), because I can't see any objects therefore it really requires an expert human judge to outline them. I attach a demo.

[mayar aly]

Sorry, i can't understand some issues. What's the next step after this step.. ok, i have detect the object, but it's only one object, what about the others !!

[Image Analyst]

I couldn't see anything in what you traced in black above, so that just confirms my thought that you need to have an expert outline these who knows what they're looking at. You could call imadjust() to try to make the contrast higher for them and make their job easier. So the next step is to just put imfreehand() into a loop. Each time after you call it you create a binary image with just that blob in it, then you OR it with the cumulative binary image of all the blobs that they're building up. Use questdlg() to ask them if they're done of have more to draw. When there are no more to draw, then you pass the binary image into bwlabel() and regionprops() to make measurements on the blobs.

Answer 2

[Image Analyst]

Try imtophat() or imbottomhat().