The BEAT platform attests that the following results were obtained by an experiment performed on our servers. We kept all the details needed to reproduce them (toolchain, database, algorithms, libraries, dataformats and the actual experimental setup).
Published 7 years, 5 months ago, on June 29, 2017, 2:52 p.m.
Experiment: sbhatta/sbhatta/iqm-face-antispoofing-test/2/replay2-antispoofing-iqm-lda
Description: Face-antispoofing expt. using image-quality measures and LDA, on ReplayAttack database
| dev_eer | 0.0633333 |
|---|---|
| dev_eer_threshold | 0.0235956 |
| dev_far | 0.06 |
| dev_frr | 0.0666667 |
| test_far | 0.055 |
| test_frr | 0.15 |
| test_hter | 0.1025 |
| dev_numNegatives | 300 |
| dev_numPositives | 60 |
| test_numNegatives | 400 |
| test_numPositives | 80 |
| dev_scoreDistribution | |
| test_scoreDistribution | |
| dev_roc | |
| test_roc |
The experiment uses 18 image-quality measures (IQM). These IQM are computed for each frame of the input video, and the feature-sets are used to construct a 2-class classifier via Linear Discriminant Analysis (LDA).
The image-quality measures used here form a subset of the measures proposed by Galbally et al:
@INPROCEEDINGS{Galbally_IEEEICPR2014_2014,
author = {Galbally, Javier and Marcel, S{\'{e}}bastien},
title = {Face Anti-spoofing Based on General Image Quality Assessment},
booktitle = {Proceedings of the 22nd International Conference on Pattern Recognition},
month = aug,
year = {2014},
}