precision and recall computation

i have a background subtracted image and its ground truth. How can i find the accuracy using precision and recall.precision=tp/(tp+fp),and recall=tp/(tp+fn). F measure=2*p*r/p+r. i want to get these values. please help

Answer 1

[Image Analyst]

How are you defining true positive (tp), false positive (fp), and false negative (fn)? Is it just based on whether the difference between your output pixel and your ground truth pixel is zero or non-zero? Wouldn't accuracy just be the ratio of true positive pixels to the number of pixels in your image?

[soumya]

i have binary image of ground truth and obtained output in binary. how should i compute tp tn etc? pls help

[Image Analyst]

If you're looking at each pixel, then count the true positives like this

% Calc true positive image, where both images are true.
match11 = (testImage == truthImage) & testImage; % This is a binary image
% Calc true negative image, where both images are false.
match00 = (testImage == truthImage) & ~testImage; % This is a binary image
% Calc false positive image, where test image is true & truth is false
match10 = testImage & ~truthImage; % This is a binary image
% Calc false negative image, where test image is false & truth is true
match01 = ~testImage & truthImage; % This is a binary image

% Count up the "true" values in the binary images.
numberOfTruePositives = sum(match11(:));
numberOfTrueNegatives = sum(match00(:));
numberOfFalsePositives = sum(match10(:));
numberOfFalseNegatives = sum(match01(:));

Does that make sense to you?

[soumya]

sir,what i need is to get the true positive of the pixels in a frame.i have a ground truth image,in which foreground region is marked white and background region are marked black.after execution of my code, i got output image.I want to check,how much of my output image matches with ground truth. is dat possible?

[Image Analyst]

Yep, that's what I gave you.

[soumya]

but sir, when i executed the code that u provided,my numberof true positives=0, where the output is mostly similar to ground truth..can u pls help?

[soumya]

probably,my two formats are different,dat could be the problem. testImage is logical,truthImage is uint8.Is dat ryt sir?

[soumya]

sir,I got result....Thank u so much sir...I was a bit nervous...sorry for the blunters.... :)

[Image Analyst]

They both need to be binary images since that is what you said at first and that's how I designed the code. If they're both gray scale images then you'll need to modify it. But whatever, they both must be the same type.

[Abdulrahman Mkaies]

 if the output image (the result image ) isnt in the same location for crosses exactly (there is an error in distance , lets say 10 pixels around the original crosses ) , Can this way adapt well ??
if not, what should I modify ?
Plzzzzz answer

[Image Analyst]

I don't have the slightest idea what you're talking about. You're not even the original poster for this message, that's a year and a half old. Please read this and start your own question.

[Abdulrahman Mkaies]

yes Sir, my question is here : http://www.mathworks.com/matlabcentral/answers/214319-accuracy-and-precision-computation , Could you give me your advise please ? , Thank you in advance