Seminar (KISON): "Disinformation detection using semi-fragile watermarking"

Program

Abstract

In this study, we propose a blind semi-fragile watermarking scheme for disinformation detection. The proposed scheme can be used for both images and videos. Fake news detection, can be proved by watermark extraction and tampered areas can be identified by a masking file stemming from the extracted watermark to provide proof of the fake content. The proposed scheme has been tested on more than 100 images, and experimental results show a high level of accuracy for fake content detection and localization with a range of attacks like copy move, vector quantization, swapface, and collage attack in the swapface. We also evaluated the scheme against unwanted manipulations, such as image compression, and the results show the same quality as with uncompressed images. In addition, we applied the mentioned attacks on the compressed images with various JPEG quality factors (QF). For a range of quality factors between 50 and 100, the normalized correlation (NC) for the extracted watermark mainly exhibits a high accuracy of more than 0.9. Furthermore, for compressed images with QF values from 30 to 50, the proposed scheme shows robustness with 80% accuracy (NC greater than 0.8). These findings confirm the efficiency and applicability of the proposed scheme.

Tanya Koohpayeharaghi

Postdoctoral researcher at KISON and Communication Networks & Social Change (CNSC) research group.