PRNU Variance Analysis for Morphed Face Image Detection

In this work, a method to detect morphed face images based on Photo Response Non-Uniformity (PRNU) is presented. More specifically, the variance of PRNU-based features across image cells is estimated to distinguish bona fide from morphed and potentially post-processed morphed face images. The proposed morph detector is shown to be robust against post-processing techniques, which are likely to be applied to conceal the morphing process, e.g. histogram equalisation or image sharpening. Tested on a database of 961 bona fide and 2,414 automatically morphed face images, a detection equal error rate (D-EER) of 10.5% is obtained over all investigated attacks, including unaltered morphed images and various post-processing techniques.