IS a classifier can be used to select a particular area from an image? How can i do it using KNN classifier?

i have clustered an image. Now i want to select a particular region of the image.

Answer 1

[Image Analyst]

OK. Try this link.

[sam CP]

How can i do it by using KNN classifier? and what is the syntax ?

[Image Analyst]

You have to have a set of reference points and a set of test points. Do you have that? I'm guessing not. So then you can't use knn yet.

[sam CP]

i didn't have that reference point. What is that reference point and how can i get that reference point?

[Image Analyst]

If you don't have a set of points that you define to be certain classes, then how can you determine if other points are close to them? I think you need to understand the concept of KNN first.

[sam CP]

Actually i just want to select the ROI by using a classifier . KNN is not a mandatory one. Instead of KNN , how can i use SVM classifier for selecting the ROI?

[Image Analyst]

You already did it one way with kmeans(). And like I said, I didn't think that was a good method.

For what it's worth, I'm attaching my KNN demo.

[sam CP]

What is the training data can be used for selecting the ROI? Is that features extracted from the manually segemented tumor region?

[Image Analyst]

You have to decide that. Maybe you say it's gray level but then everything that is the same brightness will be called a tumor simply if it's the same gray level range, regardless if it's a tumor or not.

[sam CP]

but sometimes tumor is bright and in some cases it is dark , shape is also changed in several cases

[sam CP]

u have explained in your code that, there are two classes for training .

[sam CP]

actully i have only one class , that is tumor region for training purpos ..can u explain all these more

[Image Analyst]

There are always at least two classes. In your case they would be "tumor" and "non-tumor" (or "everything else").

[sam CP]

Okay. You meant that i have to extract the features from manually segmented tumor region(ROI) and also from the non tumor region (background). Then this will be consider as the two classes and after that these classes are used for the training purpose. by using this two training classes , the KNN classifier will select the ROI and background from the test image. Can i proceed with this?

[Walter Roberson]

Yes.

You mentioned that the tumor might be brighter or darker than the rest. Be sure train on both kinds and both kinds of background.

[sam CP]

how can i extract the features of tumor section (bright or dark) and background alone , since i didn't have separated sections of both.

[Image Analyst]

I suggest you talk with the radiologist who has hired you to do this or sponsored you to get all types of images you expect to encounter. He or she may have some images.

[Walter Roberson]

You need some separated images to train with.

If you do not have that, then you need to use something like the Human Brain Atlas to do image registration against known normal images, so that you can recognize what part of the brain has been imaged and from what angle, so that you can do background subtraction, and then try to figure out whether the parts left over are significant.