Evaluate the quality of image using region-based precision and recall

How to evaluate the quality of image using precision and recall measures? I want to compare the segmented image against ground truth and generate the precision recall curve. How can I do this?

Answer 1

[Image Analyst]

I've not heard of a "precision and recall curve". Do you mean the Receiver Operating Characteristic curve? Anyway, you have to have some "ground truth" on the image - the true values or true regions or something. You can use things like ssim(), psnr(), immse(), confusion matrices, etc.

And define "quality of image". What metric or algorithm are you using to put a number on the "quality" of the image. Is it assessed on the image alone, by itself? Or is it a metric where you compare it to some ground truth image that you know the "true" answers for?

[Nataliya]

This is what i want to do

[Image Analyst]

Then just use sum() and &:

precision = sum(sum(R&Rg))/sum(R(:));
recall = sum(sum(R&Rg))/sum(Rg(:));

Where R and Rg are your two segmented binary images.