Image Segmentation - object got a blurred boundary and image contain a lot of other objects

Hello everyone, I need to detect some object/cells in an image, but the image quality doesn't good.

See the attached pictures:

Picture I need to process:

Here are the objects I am interested in:

So it should be like this, but I did this manually:

The ideal thing would be some kind of code that can detect them automatically, since I have more than 1000 images. :)

I have tried different function to detect those cells, but without any luck. It would be great with some advice to handle this problem.

Thank you for your time and help.

Best regards,

Nick

[Image Analyst]

Do you have a question? If so, what is it?

[Nick]

I have tried to apply some of the online code by using dilation, threshold and imfindcircles. However, none of these functions(at least for what I have tried) will detect the cells that I am searching for.

So my question is: What kind of function is useful to detect those cells.

Cheers, Nick

[Nick]

I have also tried to use a gaussian filter and disk on the image, but it seem like the outline of my objects/cells will be filtered away...

Answer 1

[Image Analyst]

I'd use imfreehand(). See attached demo.

[Nick]

I can tell that I didn't formulate my question that well, sorry for that.

For my problem I would like something to detect those cells automatically, since I will probably have more than 1000 images in the end.

So it should be something like shown in this MATLAB example: https://www.mathworks.com/help/images/examples/detect-and-measure-circular-objects-in-an-image.html

However, the objects are not very obvious and there are a lot of other objects that can interupt the analysis too.

Do you have a clue to find the cells without using the imfreehand?

Best regards, Nick

[Image Analyst]

I have my doubts it can be done automatically. In fact even I can't tell where the regions are.

Perhaps you can set up a competition on Kaggle.com. For example, here is one object identification competition in imagenet: https://www.kaggle.com/c/imagenet-object-localization-challenge

Or you can submit your images to a crowdsourcing site, such as Amazon Mechanical Turk http://www.toptenreviews.com/money/services/best-crowdsourcing-sites/amazon-mechanical-turk-review/

http://www.techrepublic.com/article/inside-amazons-clickworker-platform-how-half-a-million-people-are-training-ai-for-pennies-per-task/

Some comnpanies help with the Mechanical Turk process if you're unfamiliar with it. These include iKnow and and innocentive

[Nick]

Damn, I feared as much... Thank you for your time and help Image Analyst. I follow your suggestions :)

Cheers, Nick