Progress and challenges in compressible and learnable image encryption for privacy-preserving image encryption and machine learning [keynote]
Hitoshi Kiya · 2020
Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. With the wide/rapid spread of distributed systems for information processing, such as cloud computing and social networking, not only transmission but also processing is done on the intemet. However, cloud environments have some serious issues for end users, such as unauthorized access, data leaks, and privacy compromise, due to unreliability of providers and some accidents. Accordingly, we first focus on compressible image encry ption schemes, which have been proposed for encryption-then-compression (EtC) systems, although the traditional way for secure image transmission is to use a compression-then encry ption (CtE) system. EtC systems allow us to close unencrypted images to network providers, because encrypted images can be directly compressed even when the images are multiply recompressed by providers. Next, we address the issue of leamable encryption. Huge training data sets are required for machine leaming and deep leaming algorithms to obtain high performance. However, it requires large cost to collect enough training data while maintaining people's privacy.