Unauthorized AI cannot recognize me: Reversible adversarial example

In this study, we propose a new methodology to control how user’s data is recognized and used by AI via exploiting the properties of adversarial examples. For purpose, reversible example (RAE), type example. A remarkable feature RAE that image can be correctly model specified user because authorized recover original from exactly eliminating perturbation. On other hand, unauthorized models cannot recognize it functions as an Moreover, considered one encryption computer vision since reversibility guarantees decryption. To realize RAE, combine three technologies, example, hiding for exact recovery perturbation, selective AIs who remove Experimental results show proposed method achieve comparable attack ability with corresponding similar visual quality image, including white-box attacks black-box attacks.