I need to determine if one value is better significantly over others few values. I mean, I use a metric to measure different image segmentation methods, and I got these three values for that metric:

method 1: 4.23 (average of 3.06, 7.59, 2.04) method 2: 4.17 (average of 2.89, 6.67, 2.95) method 3: 2.63 (average of 1.98, 3.46, 2.45)

Is it possible to determine if the last value (method 3 = 2.63) differ significantly from the the other two methods. I understand there is few data, but it is all that I have.

I hope it was asked clear this time.

Thanks in advance, Jaime

  • $\begingroup$ Questions of statistical significance require information about the variability of the statistics you are using (your three values). Therefore you need to edit this question to include such information. $\endgroup$
    – whuber
    Dec 16, 2017 at 18:02
  • 1
    $\begingroup$ I add information about three (averaged) values. I hope it helps to be clearer. $\endgroup$ Dec 16, 2017 at 22:24

1 Answer 1


Your sample sizes are rather small, but you could cut up images into sub-parts in order to fix this issue. Perhaps visual inspection of the segmentation results will give you a hint in which sub-parts Method 3 is doing better than the others, and there you would see a greater difference in performance, if not in all such sub-parts of the image. (A sidenote, which may still be of relevance here: I don't know your segmentation method, but perhaps you could also use sub-parts of the image to choose any necessary parameters in an optimal way -- a common approach in image processing.)

  • $\begingroup$ Thanks @Cristian_Rau, I thinks your idea could work. I am going to split images to have a more information about every method's performance. I am comparing watershed and multiresolution segmentation methods versus a new method I am proposing. $\endgroup$ Dec 19, 2017 at 14:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.