AdvLUT: Cloaking Geographic Location With Semantic-Based Adversarial 3-D Lookup Tables

The proliferation of Internet of Things (IoT) devices equipped with cameras, such as those in electric vehicles, has increased the collection of personal image data. However, the potential misuse of cross-view geo-localization (CVGL) models, which can infer precise locations from ground view images, has been overlooked and seriously threatens individual location privacy. In this article, we introduce AdvLUT, a novel semantic-based adversarial 3-D lookup tables (3DLUTs) privacy protection framework designed to safeguard geographic location privacy against CVGL models. The AdvLUT employs a geographic feature encoder to extract semantic features rich in geographic information from the ground view input. These features then guide a specialized adversarial 3DLUT generator in producing a 3DLUT that alters the color properties of the input image, thereby obstructing accurate location inference. Furthermore, AdvLUT is designed with a generative architecture that enables rapid image processing within milliseconds, eliminating the need for the corresponding satellite image or CVGL model. Experimental results on multiple benchmark datasets and CVGL models demonstrate that our method achieves up to a 65.48% reduction in R@1 localization accuracy, with performance further improving to 69.25% after JPEG compression.

I. Introduction

The cross-view geo-localization (CVGL) model is designed to retrieve satellite images corresponding to ground view images. This technology offers promising applications, such as providing location information for autonomous driving systems in GPS-deprived environments or locating missing persons. However, the potential misuse of CVGL models to infer sensitive information, such as individuals’ home addresses and daily activities, has not been thoroughly explored. This issue is exacerbated by the proliferation of Internet of Things (IoT) devices equipped with cameras, such as electric vehicles and smart glasses, continuously capturing and uploading ground view images to the Internet.