eduzhai > Applied Sciences > Engineering >

CNN Detection of GAN-Generated Face Images based on Cross-Band Co-occurrences Analysis

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: Last-generation GAN models allow to generate synthetic images which arevisually indistinguishable from natural ones, raising the need to develop toolsto distinguish fake and natural images thus contributing to preserve thetrustworthiness of digital images. While modern GAN models can generate veryhigh-quality images with no visible spatial artifacts, reconstruction ofconsistent relationships among colour channels is expectedly more difficult. Inthis paper, we propose a method for distinguishing GAN-generated from naturalimages by exploiting inconsistencies among spectral bands, with specific focuson the generation of synthetic face images. Specifically, we use cross-bandco-occurrence matrices, in addition to spatial co-occurrence matrices, as inputto a CNN model, which is trained to distinguish between real and syntheticfaces. The results of our experiments confirm the goodness of our approachwhich outperforms a similar detection technique based on intra-band spatialco-occurrences only. The performance gain is particularly significant withregard to robustness against post-processing, like geometric transformations,filtering and contrast manipulations.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×