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

We propose a novel image transformation scheme using generative adversarial\nnetworks (GANs) for privacy-preserving deep neural networks (DNNs). The\nproposed scheme enables us not only to apply images without visual information\nto DNNs, but also to enhance robustness against ciphertext-only attacks (COAs)\nincluding DNN-based attacks. In this paper, the proposed transformation scheme\nis demonstrated to be able to protect visual information on plain images, and\nthe visually-protected images are directly applied to DNNs for\nprivacy-preserving image classification. Since the proposed scheme utilizes\nGANs, there is no need to manage encryption keys. In an image classification\nexperiment, we evaluate the effectiveness of the proposed scheme in terms of\nclassification accuracy and robustness against COAs.\n