Attention Aware Wavelet-based Detection of Morphed Face Images

Morphed images have exploited loopholes in the face recognition checkpoints, e.g., Credential Authentication Technology (CAT), used by Transportation Security Administration (TSA), which is a non-trivial security concern. To overcome the risks incurred due to morphed presentations, we propose a wavelet-based morph detection methodology which adopts an end-to-end trainable soft attention mechanism. Our attention-based deep neural network (DNN) focuses on the salient Regions of Interest (ROI) which have the most spatial support for morph detector decision function, i.e, morph class binary softmax output. A retrospective of morph synthesizing procedure aids us to speculate the ROI as regions around facial landmarks, particularly for the case of landmark-based morphing techniques. Moreover, our attention-based DNN is adapted to the wavelet space, where inputs of the network are coarse-to-fine spectral representations, 48 stacked wavelet sub-bands to be exact. We evaluate performance of the proposed framework using three datasets, VISAPP17, LMA, and MorGAN. In addition, as attention maps can be a robust indicator whether a probe image under investigation is genuine or counterfeit, we analyze the estimated attention maps for both a bona fide image and its corresponding morphed image. Finally, we present an ablation study on the efficacy of utilizing attention mechanism for the sake of morph detection.

1. Introduction

Robust, reliable verification systems are the crucial backbones of biometric document authentication protocols, that are to operate flawlessly. Although image morphing is not a new paradigm, it was first identified as a security concern by Ferrara et al. [8], who explained how a criminal can dodge a border control checkpoint using a travel document that was issued with a morphed image. The goal of the face image morphing attack is to synthesize a forged imaged from two composing original images such that the artificially crafted morphed image can be verified against the two original images not only visually, but also in the feature space by a classifier [28]. Moreover, morphed samples can be labeled as hard positive samples in comparison to negative genuine samples because morphed samples are synthesized to intentionally lie on the negative samples’ manifold. Similar to adversarially perturbed data samples that fool classification networks into a wrong predicted class [11], [16], morphed images are crafted to lead a verifier into a false acceptance.