Histograms of LBP features are computed for the samples in the training-set. From these histograms, mean-histograms are computed for each of the two classes: real, and attack. Given a new probe, a histograms of LBP features is computed for the probe, and this is compared with the mean-histograms of the two classes. The probe is assigned to the class closest to it, based on the chi-square distance for histogram-similarity.

This method is described in the paper by Chingovska et al:

@INPROCEEDINGS{Chingovska_IEEEBIOSIG2012_2012,
  author = {Chingovska, Ivana and Anjos, Andr{\'{e}} and Marcel, S{\'{e}}bastien},
  projects = {Idiap, TABULA RASA},
  month = sep,
  title = {On the Effectiveness of Local Binary Patterns in Face Anti-spoofing},
  booktitle = {Proceedings of the 11th International Conference of the Biometrics Special Interest Group},
  year = {2012},
  }