Content-based Graph Privacy Advisor

TL;DR

This paper introduces Graph Privacy Advisor, a simplified and more effective image privacy classifier that leverages scene and object features to better predict privacy risks of images uploaded online.

Contribution

It presents a refined graph-based model that improves privacy prediction accuracy by selecting relevant features and modeling object co-occurrences instead of simple frequency counts.

Findings

Enhanced privacy classification accuracy.

Reduced model complexity with lower-dimensional features.

Effective modeling of object co-occurrences to mitigate bias.

Abstract

People may be unaware of the privacy risks of uploading an image online. In this paper, we present Graph Privacy Advisor, an image privacy classifier that uses scene information and object cardinality as cues to predict whether an image is private. Graph Privacy Advisor simplifies a state-of-the-art graph model and improves its performance by refining the relevance of the information extracted from the image. We determine the most informative visual features to be used for the privacy classification task and reduce the complexity of the model by replacing high-dimensional image feature vectors with lower-dimensional, more effective features. We also address the problem of biased prior information by modelling object co-occurrences instead of the frequency of object occurrences in each class.