Privacy-Preserving Image Transformation Method for Person Detection and Re-ID

In recent years, the number of surveillance cameras is on the rise, and the limitation on image recognition including person re-identification (Re-ID) is lifted to ensure a safer and more convenient world. However, this has led to increasing privacy concerns. To address this issue, numerous privacy-preserving technologies have been proposed but they focused primarily on detection or classification of people. In this study, we propose a novel image transformation method to preserve privacy in detecting and re-identifying people. Privacy-preserving images visually protect images of people, but a machine can detect and classify them without any additional information. Our results show that the privacy-preserving images, transformed by our method, maintained the accuracy of person detection and Re-ID compared to that of the images with gaussian noise, blur, and pixelation.