I have two matrices corresponding to genuine & imposter representing hamming distance. Now i want to calculate FAR, FRR and efficiency. how i can calculate...
Maybe it is a first step to define some threshold (a Hamming distance below or above that threshold then corresponds to rejection or acceptance of the subject under consideration). With the FAR and stuff you can build up an ROC-curve that gives good overview of the "efficiency" of your biometric system. To build up that ROC-curve you can, for example, sweep the mentioned threshold. A very interesting value in this context is the EER (Equal Error Rate). A very good textbook on this topic is "Biometric System and Data Analysis Design, Evaluation, and Data Mining" by Ted Dunstone and Neil Yager. Have a look in chapter 2, please.
http://en.wikipedia.org/wiki/Receiver_operating_characteristic
http://gim.unmc.edu/dxtests/roc2.htm
Before one speaks of thresholds and Hamming distances it's important to finally agree to change terminologies. FRR (Type I error) is an IMPOSTER.( especiallyif used in a deliberate way) FAR (Type II error) is an IMPERSONATOR.. The term 'imposter" for both is incorrect .
Leonard, I upvoted your post for it seems very interesting (and correct and important) to me. Thank you!
But can you please give an example on where the "effective" difference between the terms imposter and an impersonator lies?
To me, members of both groups (imposters and impersonators) effectively do the same: they both inject type II errors (FAR) into the verification system, given they do their job successful - for both try to counterfeit some biometric data (iris, fingerprints, signatures, whatever).
A type I error is induced into the system by someone who does not want to counterfeit, but gets rejected due to some inabilities to match own templates, so a false reject occurs.
So in my opinion a member of the individuals who induce type II errors is neither an imposter nor an impersonator, but someone who should be called "poor sod" or the like, for not being able to reproduce biometric data that can be matched correctly.
Maybe this is a problem only for me because I am not a native English speaker. I had a look into my dictionary, and both terms become translated to something like "cheater".
Maybe you can spare a moment on this. Thank you in advance!
Benjamin
PS and to tell a long story short: Of course, imposter for both types of errors is clearly incorrect, as mentioned by Leonard.
- Badges
- Science topic
- Similar topics
- Books