The prevalence of facial recognition, biometric unlock, and social media presents a significant opportunity for bad actors to introduce forged or manipulated images to spread false information or damage reputations. This is aided by the continuing improvement in realistic image synthesis and manipulation by generative adversarial network, GAN, based methods. We introduce a method for the detection of forged images and localization of manipulated regions in face images.

Training architecture for the proposed approach.

Figure 1: The human face is under attack from three directions. Here we fight the third type, digitally manipulated or synthesized images.

To simplify the localization of manipulated regions, we develop the manipulation appearance model, MAM. This utilizes basis templates for commonly manipulated regions. Instead of estimating the entire localization map, MAM can predict a 10-d vector that is combined with the templates to produce a manipulation mask for the image.

Illustration of the template bases in the MAM model.

Figure 2: Illustration of the template bases in the MAM model.

DFFD Dataset

DFFD dataset download page: http://cvlab.cse.msu.edu/dffd-dataset.html

Detection of Digitally Manipulated Faces Source Code

The source code for the networks is available at: https://github.com/JStehouwer/FFD_CVPR2020

Publications

  • On the Detection of Digital Face Manipulation
    Hao Dang*, Feng Liu*, Joel Stehouwer*, Xiaoming Liu, Anil Jain
    In Proceeding of IEEE Computer Vision and Pattern Recognition (CVPR 2020), Seattle, WA, Jun. 2020
    Bibtex | PDF | arXiv
  • @inproceedings{ on-the-detection-of-digital-face-manipulation,
      author = { Hao Dang* and Feng Liu* and Joel Stehouwer* and Xiaoming Liu and Anil Jain },
      title = { On the Detection of Digital Face Manipulation },
      booktitle = { In Proceeding of IEEE Computer Vision and Pattern Recognition },
      address = { Seattle, WA },
      month = { June },
      year = { 2020 },
    }