How to perform image segmentation

I want to manually segment a small object(eg: a feature similar to cell nucleus) which is located within the original image. Can anyone kindly tell me what is the cause to obtain the boundary of the segmented image larger than the real image? (though i drew the boundary along the cell nucleus, segmented image still gives larger segmentation.

Thank you.

Answer 1

[Image Analyst]

See my Image Segmentation Tutorial : http://www.mathworks.com/matlabcentral/fileexchange/25157-image-segmentation-tutorial

[Thomas Koelen]

I don't think this is the issue, his problem is that his cut outs include the backround, which is still the case in your tutorial, maybe try to explain him why you can't cut out "just" an object from an image.

[Image Analyst]

Like Thomas said, images must remain rectangular. Attaching your image would be more illustrative than making us work "blind". I'm assuming you used imfreehand() so show us screenshots of what you did and what you ended up with. For what it's worth, I attach 3 of my imfreehand demos.

[Chathu]

@ Image Analyst: i am really sorry for not replying you quick. Let me have a look at your code and examples and see whether i can figure out from that. Image Analyst and Thomas- thanks a million for your support, so far. Really appreciate your time and effort on this.

[Image Analyst]

From your image in your so called "Answer" below, it looks like you could just take the red channel and threshold it to find dark things:

binaryImage = rgbImage(:,:,1) < someThresholdValue;
labeledImage = bwlabel(binaryImage);
measurements = regionprops(labeledImage, 'Area', 'Centroid');

There are other things you could measure but I don't know what you want to know about that black speck.

[Chathu]

@ Image Analyst, thank you very much for your response.

[Image Analyst]

Chathu, that's what my Image Segmentation Tutorial does. http://www.mathworks.com/matlabcentral/fileexchange/25157-image-segmentation-tutorial Just tell it to get the dark things instead of the bright things. It will crop them out into separate subimages. Then you can just save the subimages with imwrite().

[Chathu]

Image Analyst- thank you so much for the tutorial and your responses. Really admire your effort.

Answer 2

[Thomas Koelen]

If I understand you correctly, you want to extract only the nucleus, and not the background, this is impssible because you can't make an image that's not a rectangle. You can however make the backround white.

[Chathu]

Thomas: please excuse me for replying you late. But yes i understand your point. I am still trying to figure it out.

Answer 3

[Chathu]

@ Image Analyst and Thomas- i was trying hard to segment this (pls see the attachment- arrow indicates the portion that i need to segment). Can anyone of you kindly tell me what is the best way/method to segment this?

[Thomas Koelen]

Since the part you are trying to get is way darker than the other parts I suggest you can do something like this:

(im using a paint image because you didn't supply your own image wihout the arrow)

IM=imread('chathu.png');
figure
IM(IM>30)=255;
bw1=im2bw(IM);
imshow(bw1);
IM2=medfilt2(bw1);
stats = regionprops(IM2,'Centroid');

which give soyu the center of the "blob":

stats =

    Centroid: [134.3417 148.9362]

[Chathu]

@ Image Analyst,

@ Thomas,

Thank you so much for both of your responses. Really appreciate it. Please kindly note that i need to segment(extract) the black speck in a separate window. Since i have huge set of black specs like this, my intention is to segment all black specs, save them in a separate folder, then stack them and finally, model it.