Roc curve for image segmentation

Hello !

I have been trying to draw ROC curve for my segmented image ...

I have Ground truth image and my segmented image ...

Now i want to draw ROC curve ,,i have gone through many threads but i couldnot understand the basic theme of TRue positive,false positive,false negative and true negative.... ..

I have Dice similarity coefficient of say 5 images

0.94 0.77 0.85 0.89 0.93

and ground truth images

Is there any code or suggestion on how to calculate different rates and draw ROC ???

What i read on google,most of the threads says that ROC curve required a classifier ..but in my case there is no classifier ..so what to do ?

Also how to calcualte the pixels which are not a part of the background or object ???

[Image Analyst]

Please attach your ground truth segmentation image and your "test" image segmentation. Where they match is a true positive or true negative, and where they don't match is a false positive or false negative. And how many classes do you have? Just two - foreground and background - or do you have more?

[jyothi B]

Answer 1

[Thorsten]

For a ROC curve you need a binary ground truth and a continuous-valued segmented images; this image is usually the output of an operator or classifier that you've run on the original image. Then you threshold the continuos-valued output image of the classifier at various values, compute true-positive and false-positives for each threshold, and finally plot tp against fp, starting with the highest threshold.

If you have a binary ground truth and a binary segmented image, you cannot compute an ROC.

[mahreen kiran]

Ok i got that point

 I have binary ground truth image and binary segmented image ..Can u please suggest some other method that can be used as equivalent to ROC ????

[mahreen kiran]

Also i wanna know if it possible to use my similarity measure (which are overlapping scores like 98.3 ,89.7 etc ) as a continous level and set my threshold to 70 ... if overlapping score is less than 70 than it would be 0 otherwise 1 ...

 would it be acceptable?

[Thorsten]

As I said, you cannot use ROC if both images are binary.

[Nataliya]

Hi Thorsten, I want to generate precision recall curve. I have binary ground truth and continuous valued segmented image. Please guide me how can I do this. Can you help me get the code please? Here is my question: https://www.mathworks.com/matlabcentral/answers/316327-evaluate-the-quality-of-image-using-region-based-precision-and-recall