0

I am working on a vein pattern recognition project based on SURF algorithm and euclidean distance. I have completed my program to find the maximum and minimum distance between vein features and find a match exactly when there is an identical image. i.e max and min distance between two images is zero. In this case, how would I find my FAR and FRR. Will it be 0% or am I missing a big concept here?

Even if there is a slight variation it wouldn't match in which case, I guess I need to have a threshold value to compare to. I have calculate the max and min distance between all combination of images with the same hand, with different hands. In this case, how do I computer the FAR and FRR. This is my first biometrics project and it would be helpful if I am directed to any resource that would help me in this. Thank you.

Kindly help me out.

Srikkanth Govindaraajan
  • 187
  • 3
  • 18

1 Answers1

0

Usually, it's impossible to find a model int the training set that exactly matches a sample in the testing set. (The situation that the euclidean distance is zero is impossible in the reality.) So we use many kinds of similarity measure methods, such as in your case, use euclidean distance.

And then we should have a threshold to determine whether a test sample and a certain model in training set is belong to the same person. If the distance is larger than the threshold, we reject the hypothesis that they are the same person and accept otherwise. So we have false reject and false accept cases.
In your case, maybe you can avoid choosing a threshold as every test sample is compare to all the training models. So you can just choose the nearest training model as your recognition result.

To calculate FAR and FRR, you should have the key labels to your test samples so that you can compare you recognition results with the keys.

Bill Xia
  • 299
  • 2
  • 14