Tampering detection and localization in images from social networks: A CBIR approach

Abstract

Verifying the authenticity of an image on social networks is crucial to limit the dissemination of false information. In this paper, we propose a system that provides information about tampering localization on such images, in order to help either the user or automatic methods to discriminate truth from falsehood. These images may be subjected to a large number of possible forgeries, which calls for the use of generic methods. Image forensics methods based on local features proved to be effective for the specific case of copy-move forgery. By taking advantage of the number of images available on the internet, we propose a generic system based on image retrieval, followed by image comparison based on local features to localize any kind of tampering in images from social networks. We also propose a large and challenging adapted database of real case images for evaluation.