Alaska

A challenge of steganalysis into the wilderness
of the real world

About Alaska

Since more than a decade, researchers all around the world have been striven to find novel methods that allows hiding data into innocuous-looking objects which are often digital media: videos, sounds, images ; most frequently images. This is often referred to as "modern steganography". This also gives birth to another research domain usually referred to as "modern steganalysis". Indeed, in while steganography was the focus of many researches, researchers have also developed various methods for detecting objects, and especially media, that contains hidden data.

ALASKA is the second contest on steganalysis ; after a fruitful first contest, called BOSS and organized in 2010, which give birth to the developement of large feature sets and dedicated decision methods that have been extended and used for detection of steganography in images, both uncompressed and compress using the JPEG standard, but also for forensics to give a few examples. The BOSS contest has also been a great success for providing a common and reference dataset to the community.

So if BOSS contest has been a great success, why do we need a second contest ? Indeed, BOSS contest greatly help the community, however, we believe that after 8 years, the field has significantly changed, with the introduction of deep learning methods and with the improvement of embedding techniques, and that it is time to organize a second competition to benchmark current research in steganalysis. Besides, we believe that BOSS dataset does not faithfully reflect the high diversity of media that can be found "in the real world". Indeed those media are rather homogeneous since they are all processed from raw files in the same way, they are all uncompressed, in grayscale color, with the same size, etc.

ALASKA, why such a strange name for a contest on steganalysis ? The origin of the present contest finds its root in recent work that aims at detecting hidden data in highly heterogeneous images, that would reflect what can be found when inspecting images on the Internet. Surprisingly, we have faced tremendous difficulties on such dataset and found that the results obtained using the BOSS dataset are not in agreement with what would be found in practical applications. We have come up with the name "steganalysis into the wild" to describe the high diversity of media from the Internet and, referring to the movie and the book "into the wild" in which the hero spend a while in the state of Alaska. The present ALASKA contest aims at pushing the community to study the practical problems of pushing steganalysis "into the wild", with highly heterogenous color images, and we have naturally decided to use this name for this contest and match it with Application on Large and heterogeneous dataset of Steganalysis techniK for advances into the wild.

Because we believe that the results of the ALASKA contest are interesting for the whole community, the methods used by the three competitors who achieve the best performances will be presented at the ACM conference on Information Hiding and Multimedia Security (IH&MMSec) help in Paris in June 2019 while other competitors whose would like to prensent innovative works will also be welcome to submit a paper. A special session will be dedicated to the description of ALASKA contest organization, and presentation from the selected competitors.

Timeline

Material

You must be authenticated.

Rules

The contest will take place from 1st of September 2018 to 28th of February 2019, lasting 6 months in total. The rules are the following :
  • Competitors cannot submit more than one trial every four hours.
  • Competitors must create an account to be able to participate and to download any material. The accounts will only be used for statistical purposes, to contact you in case of important issue, and to ensure that the limit of one submission every four hours.
  • Each submission is evaluated over a randomly selected subset of 80% of the testing set. The final results, when contest closes, will be adjusted with evaluation over the whole testing set.
  • The ranking is made using the empirical probability of missed detection for a fixed empirical probability of false alarm of 5%. We will count of many images with hidden data are incorrectly classified as covers when exactly 5% of cover images are incorrectly classified as containing hidden data.
  • Competitors who achieve the three best performance, based on the above criterion, and those who are willing to present innovative methods, will be given the opportuinity to submit a paper for IH&MMSec 2019 conference, help in Paris, such that they will have a chance to present their works to the community.
  • Each submission is made by submitting an answer containing the image numbers ordered from the most likely containing hidden data to the less likely one. This will allow us to process submission to compute false alarm and missed detection probabilities when adjusting the threshold ; typically to draw ROC curves.
  • An example of valid submission file is provided in the submit an answer section.
  • During the warm-up phase #1, the answer consists in 200 indices with the firsts hundred corresponding to cover and the lasts hundred corresponding to stego media.

Submit an answer

You must be authenticated.

LeaderBoard

As of date (dynamique) the current leaderbord is the following. Note that you can click on column headers to sort submissions according to the various scores. However the one used for the "official" ranking is only the first one, MD005 (Missed detection for False alarm of 0.05).

Stage 1
Stage 2
Stage 3
Submission Date
Missed Detection at 5% False Alarm Rate
minPE (Minimal Error Rate)
FP50 (False Positive rate at 50% Missed Detection)

Acknowledgements

The ALASKA contest has been, in part, inspired by the BOSS competition and has been jointly and proudly organized by:
  • Rémi Cogranne, from Troyes University of Technology.
  • Quentin Giboulot, from Troyes University of Technology (PhD student of Rémi Cogranne, who did not really choose to be part of the organization but enjoyed it).
  • Patrick Bas, from École central de Lille.
We would like to thank all the individuals that help us organizing this contest. Those are mainly (but not exclusively):
  • Antoine Prudhomme, for creating the present website.
  • Julien Flamant, Jean-Baptiste Gobin, Florent Pergoud, Luc Rodrigues and Emile Touron for kindly provide some of their raw images, as well as Florent Retraint, for the sharing of ressources, within cybersec platform, under his responsibility.
  • The computer resources department of Troyes University of Technology who helped us with all their advices and suggestions.