Title:Privacy-Preserving Deep Neural Networks with Pixel-based Image Encryption Considering Data Augmentation in the Encrypted Domain

Abstract:We present a novel privacy-preserving scheme for deep neural networks (DNNs) that enables us not to only apply images without visual information to DNNs for both training and testing but to also consider data augmentation in the encrypted domain for the first time. In this paper, a novel pixel-based image encryption method is first proposed for privacy-preserving DNNs. In addition, a novel adaptation network is considered that reduces the influence of image encryption. In an experiment, the proposed method is applied to a well-known network, ResNet-18, for image classification. The experimental results demonstrate that conventional privacy-preserving machine learning methods including the state-of-the-arts cannot be applied to data augmentation in the encrypted domain and that the proposed method outperforms them in terms of classification accuracy.