A GAN-Based Image Transformation Scheme for Privacy-Preserving Deep Neural Networks
This addresses privacy concerns in image processing for users of deep neural networks, but it is incremental as it builds on existing GAN and encryption methods.
The paper tackles the problem of privacy-preserving image classification by proposing a GAN-based image transformation scheme that protects visual information without encryption keys, achieving results such as enhanced robustness against ciphertext-only attacks and maintaining classification accuracy in experiments.
We propose a novel image transformation scheme using generative adversarial networks (GANs) for privacy-preserving deep neural networks (DNNs). The proposed scheme enables us not only to apply images without visual information to DNNs, but also to enhance robustness against ciphertext-only attacks (COAs) including DNN-based attacks. In this paper, the proposed transformation scheme is demonstrated to be able to protect visual information on plain images, and the visually-protected images are directly applied to DNNs for privacy-preserving image classification. Since the proposed scheme utilizes GANs, there is no need to manage encryption keys. In an image classification experiment, we evaluate the effectiveness of the proposed scheme in terms of classification accuracy and robustness against COAs.