Home | Resume | My calendar | PublicationsDemoLIRMMBase | Job opportunities | Lecture support | Contact


LIRMMBase

LIRMMBase is indeed made of three databases, and was created in June 2015: 

- LIRMMBaseColor: A color database obtained thanks to original uncompressed RAW images (ufraw was used for the demosaicking; see the script). The database is given there: LIRMMBaseColor_part1, LIRMMBase Color_part2.
  
- LIRMMBase512x512 (grey-levels unselected images): The grey-level version of LIRMMBase Color (see the script for conversion, resizing, and cropping). The database is given there: LIRMMBase512x512.

- LIRMMBase256x256 (grey-levels selected images): The cropped version of LIRMMBase 512x512 with a selection of images and cameras. The database is given there: LIRMMBase256x256.


The initial purpose of this new database was to make some tests in the case of cover source mismatch in steganography / steganalysis. This database is to be used conjointly with the BOSSBase. The images come from several database: Columbia, Dresden, Photex, and Raise databases.

The origin of the images with the brands and the models of cameras, for LIRMMBaseColor, and LIRMMBase512x512, are listed in Table 1.

The selected images with the brands and the models of cameras, for the LIRMMBase256x256, are listed in Table 2.

Brand Model Source # of images
Nikon D40 RAISE 76
Nikon D90 RAISE 2 276
Nikon D7000 RAISE 5 804
Canon XT Columbia 117
Nikon D70s Dresden 361
Nikon D200 Dresden 752
Vosskuhler CCD 1300LN Photex 5 934
Table 1. Origin of the images with the brands, and the models of the cameras, for LIRMMBaseColor, and LIRMMBase512x512.

Brand Model Source # of images
Nikon D40 RAISE 168
Nikon D90 RAISE 168
Nikon D7000 RAISE 168
Canon XT Columbia 168
Nikon D70s Dresden 168
Nikon D200 Dresden 168
Table 2. Origin of the images with the brands, and the models of the cameras, for LIRMMBase256x256.

The script (script) we used is the same as the one used for BOSS. In order to obtain the LIRMMBase256x256, first, the script converts all the images in ppm (LIRMMBaseColor). Then, the images are first resized so that the smaller side is 512 pixels long, then images are cropped to 512x512, and finally converted to greyscale (LIRMMBase512x512). Finally we selected from 6 cameras listed in Table 1, only 42 images. The obtained  LIRMMBase256x256 database is thus made of 1008 grey levels images, coded on 8 bits, and of size 256x256. There are 6 cameras and 168 images per camera.

HOW TO CITE the LIRMMBase database:

If researchers use this free-access database, they are required to cite it as follows:
" LIRMMBase: A database built from a mix of Columbia, Dresden, Photex, and Raise databases, and whose images do not come from the same cameras as the BOSSBase database ; For the first time used in [1] ", L. Pibre, J. Pasquet, D. Ienco, and M. Chaumont, LIRMM Laboratory, Montpellier, France, June 2015, Website: www.lirmm.fr/~chaumont/LIRMMBase.

[1] Lionel Pibre, Jérôme Pasquet, Dino Ienco, and Marc Chaumont, "Deep Learning for steganalysis is better than a Rich Model with an Ensemble Classifier, and is natively robust to the cover source-mismatch," in Proceedings of Media Watermarking, Security, and Forensics, Part of IS&T International Symposium on Electronic Imaging, EI'2016,  San Francisco, California, USA, 14-18 Feb. 2016, 10 pages. (pdf).
FAQ and parameters are given here: SteganalysisWithDeepLearning.html.