You are here

Image Forensics in Social Media

Team and supervisors
Department / Team: 
Team Web Site: 
https://www.inria.fr/equipes/linkmedia
PhD Director
FURON Teddy
Co-director(s), co-supervisor(s)
KIJAK Ewa
Contact(s)
NameEmail addressPhone Number
KIJAK Ewa
ewa.kijak@irisa.fr
02 99 84 75 55
PhD subject
Abstract

Nowadays images are widely spread through social networks and the dissemination of false information is a scourge. In particular, modifying image content is an increasingly popular practice thanks to the various user-friendly softwares making image editions very easy. Tampering detection and localization in images has drawn a lot of attention. Passive, blind image forensics, based on single image analysis, aim at identify local discrepancies that can be revealed by various types of information: traces left by JPEG compression [10,11,12], Color Filter Array (CFA) interpolation [13], PRNU analysis [8] or noise patterns [9].

However, images collected from the web or social media have particular characteristics. Users often modify an image before re-posting it, and some image publishing platforms, e.g. Facebook or Twitter, automatically operate such transformations, e.g. scaling, changing format to JPEG, varying quality, and erasing metadata. Several recent datasets have been proposed to challenge these approaches [4,7], and we already know that most algorithms failed to detect any trace, most likely because the traces sought are too fragile to survive the typical image processing operations such images undergo. The vast majority of algorithms proposed for forgery detection are, by their definition, not designed to work on images that have been re-compressed or, even worse, resampled due to scaling [5,1].

One recent fruitful direction was the combination of image forensics algorithms with a reverse image search engine. This family of splicing detection algorithms attempts to take advantage of the fact that forgeries found on images from social media are often composed from source images that can also be found on the Web [1,2,4,6]. Of course, we can never exclude the possibility that no source image can be located.

In the Linkmedia team, we started to study the tampering localization in images from social media, based on local features and image retrieval techniques, and we collected specific datasets for evaluation [4]. We found out that many degenerate cases arise, calling for appropriate treatments.

In the context of images from social media, the goal of this thesis is to tackle the problems of:
* Provenance Filtering: Given a probe image, identify images from the world data set which contributed to creating the probe image.
* Splice Detection and Localization: Given several images, detect if a region of a donor image has been spliced into a probe image.
* Image Manipulation Detection and Localization: Given a single probe image, detect if the probe was manipulated.

Some related issues are the identification of the original image (related to phylogeny [3]), the characterization of non-malicious modifications, and extension to videos.
The involved techniques obviously cover the analysis and description of images, the near-duplicate retrieval, the image forensics tools, as well as learning methods [14]. In particular, it will be investigated how deep learning can apply to the forgery localization (and not only their detection [15]), in images from social networks.

Bibliography

[1] Joel Brogan, Paolo Bestagini, Aparna Bharati, Allan Pinto, Daniel Moreira, Kevin Bowyer, Patrick Flynn, Anderson Rocha, and Walter Scheirer, "Spotting the difference: Context retrieval and analysis for improved forgery detection and localization". In International Conference in Image Processing (ICIP), 2017.

[2] L. Gaborini, P. Bestagini, S. Milani, M. Tagliasacchi and S. Tubaro, "Multi-Clue Image Tampering Localization," 2014 IEEE International Workshop on Information Forensics and Security (WIFS), pp. 125-130, 2014.

[3] A. A. de Oliveira et al., "Multiple Parenting Phylogeny Relationships in Digital Images," in IEEE Transactions on Information Forensics and Security, vol. 11, no. 2, pp. 328-343, Feb. 2016.

[4] C. Maigrot, E. Kijak, R. Sicre, V. Claveau. "Tampering detection and localization in images from social networks: A CBIR approach", International Conference on Image Analysis and Processing (ICIAP), 2017

[5] Markos Zampoglou, Symeon Papadopoulos, and Yiannis Kompatsiaris. "Large-scale evaluation of splicing localization algorithms for web images". Multimedia Tools and Applications, 76(4):4801–4834, 2017

[6] I. Amerini, R. Becarelli, R. Caldelli, and M. Casini, “A feature based forensic procedure for splicing forgeries detection,” Mathematical Problems in Engineering, vol. 2015, 2015.

[7] M. Zampoglou, S. Papadopoulos, and Y. Kompatsiaris, “Detecting image splicing in the wild (WEB),” in 2015 IEEE International Conference on Multimedia & Expo Workshops(ICMEW), pp. 1–6, 2015.

[8] Davide Cozzolino, Francesco Marra, Giovanni Poggi, Carlo Sansone, Luisa Verdoliva, "PRNU-Based Forgery Localization in a Blind Scenario", International Conference on Image Analysis and Processing (ICIAP), pp 569-579, 2017

[9] D. Cozzolino, G. Poggi, and L. Verdoliva, “Splicebuster: A new blind image splicing detector,” in IEEE Int. Workshop on Information Forensics and Security (WIFS), pp. 1–6, 2015.

[10] T. Bianchi and A. Piva, “Image forgery localization via block-grained analysis of jpeg artifacts,” IEEE Trans. on Information Forensics and Security, vol. 7, no. 3, pp. 1003–1017, 2012.

[11] I. Amerini, R. Becarelli, R. Caldelli, and A. Del Mastio, “Splicing forgeries localization through the use of first digit features,” in IEEE Int. Workshop on Information Forensics and Security (WIFS), 2014, pp. 143–148.

[12] H. Farid, “Exposing digital forgeries from jpeg ghosts,” IEEE Transactions on Information Forensics and Security, vol. 4, no. 1, pp. 154–160, 2009.

[13] P. Ferrara, T. Bianchi, A. D. Rosa, and A. Piva, “Image forgery localization via fine-grained analysis of cfa artifacts,” IEEE Trans. on Information Forensics and Security, vol. 7, no. 5, pp. 1566–1577, 2012.

[14] D. Cozzolino and L. Verdoliva, "Single-image splicing localization through autoencoder-based anomaly detection," IEEE International Workshop on Information Forensics and Security (WIFS), 2016.

[15] Y. Rao and J. Ni, "A deep learning approach to detection of splicing and copy-move forgeries in images," 2016 IEEE International Workshop on Information Forensics and Security (WIFS), 2016.

Work start date: 
01/09/2018
Keywords: 
Image forensics, false information detection, social media, forgery localization, deep learning, phylogeny
Place: 
IRISA - Campus universitaire de Beaulieu, Rennes