eduzhai > Applied Sciences > Engineering >

A GAN-Based Image Transformation Scheme for Privacy-Preserving Deep Neural Networks

  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: We propose a novel image transformation scheme using generative adversarialnetworks (GANs) for privacy-preserving deep neural networks (DNNs). Theproposed scheme enables us not only to apply images without visual informationto DNNs, but also to enhance robustness against ciphertext-only attacks (COAs)including DNN-based attacks. In this paper, the proposed transformation schemeis demonstrated to be able to protect visual information on plain images, andthe visually-protected images are directly applied to DNNs forprivacy-preserving image classification. Since the proposed scheme utilizesGANs, there is no need to manage encryption keys. In an image classificationexperiment, we evaluate the effectiveness of the proposed scheme in terms ofclassification accuracy and robustness against COAs.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×