Securing face recognition systems against replay attacks has been recognized as a real challenge. In this work, the problem of fake face detection is addressed by modelling radiometric distortions involved in the recapturing process. The originality of our approach is that the fake face detection process occurs after the face identification process. Having access to enrolment data of each client, it becomes possible to estimate the exposure transformation between a test sample and its enrolment counterpart. A compact parametric representation is proposed to model those radiometric transforms and is used as features for classification. We evaluate the proposed method on Replay-Attack, CASIA and MSU public databases and prove that our method is competitive with state of the art countermeasures.