PPUP-GAN: A GAN-based privacy-protecting method for aerial photography

With the rapid development of unmanned aerial vehicles (UAVs), aerial photography has also emerged, which brings potential risk of privacy leakage. It is necessary to balance the privacy protection and utility of aerial photography data in UAVs. To address the issue, a GAN-based privacy-protecting and utility-preserving method called PPUP-GAN is proposed. In the PPUP-GAN, the object of interest and background of aerial photography can be defined and processed separately. Then, the balance of privacy-protecting and utility-preserving can be innovatively achieved by maintaining the content of the objects of interest while changing the content of the privacy-related background. Specifically, the image-to-image transformation method is adopted to reduce the risk of privacy leakage of background. In this way, the processed aerial photography can still be employed for subsequent image analysis. We evaluate our method on a real-world UAV image dataset. Experimental results show that our methods can greatly reduce the probability of bystanders being detected in the background. And the security analysis shows that the PPUP-GAN has the ability to resist model inversion attacks.(c) 2023 Elsevier B.V. All rights reserved.