I am developing EEG based biometric authentication system. I extracted features from 10 subjects' of EEG data. Now I want to measure the performance of my system. I want to consider the features as a template and match them during authentication. What should be the correct evaluation technique of my system and what are the formulas of those evaluation metrics?
Biometric authentication performance is generally shown in 2 ways:
1. ROC (along with AUC) + EER + Classification accuracy at particular FAR.
2. CMC (+Rank-wise matching performance)
For first, it is a verification set up. You report pair-wise matching as a match(1)/non-match(0). Using this, you report performance using the ROC curve and related metrics. These metrics include area under the curve (AUC), equal error rate (EER), and Classification accuracy at particular FAR..
For the second, you match each query (probe) sample with all templates (gallery) and report a cumulative matching characteristic curve (CMC). It shows rank-wise matching accuracy, additionally reporting Rank-1 and Rank-k accuracy (k can be 5, 10, 50 depending on the number of subjects) in a tabular format too.
Decide either of the two as per the common norms followed in the literature for your problem.
- Badges
- Science topic