AlphaNet: Single Morphing Attack Detection Using Multiple Contributors

This paper proposes and explores a Single Morphing Attack Detection (S-MAD) method of morphed face images created from different numbers of contributors. We selected facial images from K subjects and their accomplices for developing morphing face images from K = 2, 4, 8 and 16 subjects. Morphing images with more contributors appear as smothered faces. We developed a multiclass Convolutional Neural Network (CNN) based on three levels of alpha filters applied to the RGB channels based on MobileNetV3 (AlphaNet). This AlphaNet was evaluated on intra-dataset and cross-dataset scenarios. Our method tested in synthetic images obtained an BPCER
10
of 4.41 % and BPCER
20
of 4.56% on cross-dataset.