------------ StirMark 3.1 ------------ Fabien A. P. Petitcolas and Markus G. Kuhn University of Cambridge, Computer Laboratory Broad claims have been made about the `robustness' of various digital watermarking method. Unfortunately the criteria as well as the pictures used to demonstrate these claims vary from one system to the other and recent attacks show that the robustness criteria used until now are inadequate: JPEG compression, additive Gaussian noise, low pass filtering rescaling and cropping have been addressed in most the literature but specific distortions such as rotation have been rarely addressed. In some cases the watermark is simply said to be ``robust against common signal processing algorithms and geometric distortions when used on some standard images.'' Most of the potential attacks detailed in: Martin Kutter and Fabien A. P. Petitcolas. A Fair Benchmark for Image Watermarking Systems. To be presented at the 11th Annual Symposium on Electronic Imaging, IS&T/SPIE, San Jose, USA, 23-29 January 1999. are actually implemented into this version of StirMark: given a watermarked image, StirMark will apply these transformations with various parameters. Then the output images can be tested with watermark detection or extraction programs. The full process can be atomised using a simple batch file. For this benchmark we suggest to use pictures freely available for research purpose and downloadable from: A result sheet for this benchmark applied to various copyright marking algorithms can be download from: --- The original StirMark (version 1.0) is still available from --- StirMark reads and writes arbitrary images in the PGM, PPM and JPEG file formats. The selection is based on the extension of the file only: .ppm, .pgm, .jpg or .jpeg ---------------------------- What's new with version 3.1? ---------------------------- New features in version 3.1: - Possibility to select a category of tests only (e.g. rotations only). - New tests: general linear geometric distortions, change of aspect ration, different sizes for median filtering, asymmetric row and column removal, new default values for some existing tests (especially. JPEG compression). - Option to save and re-use parameters of geometrical distortions. --- Version 3.0 adds: - New transformations including row and columns removal, shearing in X and Y direction and sharpening. - Color quantisation (used by GIF) support based on Jef Poskanzer's code. - The Laplacian based removal attack using Richard Barnett's is now applied before the geometrical distortions. - Basic quality measure functions: SNR and PSNR. - Better JPEG support for input image including simple autodection of the input format based on file extension (pgm, ppm, jpg or jpeg). --- Version 2.3 now uses the bi-quadratic resampling algorithm instead of the Nyquist interpolation algorithm used in previous version. The speed improvement is roughly a factor 8 to 9 for a 700x500 image. The bi-quadratic algorithm gives a result similar to the bi-cubic in about 60% of the CPU time. It is described into: Neil Dodgson, "Quadratic Interpolation for Image Resampling," IEEE Transactions on Image Processing, vol. 6, no. 9, pp. 1322--1326, Sept. 1997. --- Version 2.2 adds the following command line option to version 1.0 -T to perform a standard series of tests on a single image. -R and -b are used in the enhanced distortion process. These options are described in details below. Version 2.2 now uses the JPEG library of the Independent JPEG Group's free JPEG. For convenience this library is provided in the StirMark package but the latest version can be downloaded from: So the default output now uses the JPEG format: stirmark test.ppm output.jpg just applies the latest StirMark distortions to the image, including JPEG compression at the end. -------------------- Command line options -------------------- You can specify an input and an output file name in the command line. If you omit any of those, stdin and stdout are used instead respectively. The output uses the JPEG format. -? To get help. -b The number of pixel displacement allowed for the center of the image. All the pixel are moved using a smooth function. Pixels at the corners are not moved while the pixel in the center of the image is moved the most. This is similar to a `punch/pinch' effect. -d The maximum byte value by which any of the RGB values of any pixel is allowed to deviate from the original value (default 1.5). -i The number of pixel distances that the corner of the target image is allowed to be inside the original image. If the number is followed by %, then the maximum corner shift is defined as a percentage of the image width or height, whichever is smaller. Default is -i2.0%. -ll This option is specific to the frequency mode Laplacian removal attack which is applied before the random geometric distortions. It is used to specify the LR strengths for scale 2^n. For more details, see: -lj This option is also specific to the FMLR attack. It is used to specify the JND LR strengths for scale 2^n. For more details, see: -o The number of pixel distances that the corner of the target image is allowed to be outside the original image (default 0.7). Sample values taken from outside the original image are extrapolated. You can also specify a percentage as with -i, but this is normally not useful as -o values much higher than 1 cause mostly useless extrapolations. -PS Save the main distortion parameters into a file. These parameters include those specified from the command line (or default if not specified) as well as other random parameters. Useful is you want to apply the same transformation to a sequence of frames in a video. This avoid getting shaking effect. -PS+ Save the main distortion parameters and all the random numbers used the in the geometrical distortion process. This creates very big files. -PL Load previously saved main random parameters using -PS or -PS+ options. -PL+ Load previously all previously saved parameters (using -PS+ option). If -PL+ is used after -PS, StirMark switches it back automatically to -PL. -q JPEG quality factor used after geometric distortions. -NOJPEG Prevent StirMark from applying JPEG compression after the geometrical distortions. -R The fraction of pixel displacement allowed for any pixel. During the distortion process, any pixel can be moved randomly along both axes. -s Seed value for the random number generator. If you do systematic measurements with StirMark, then repeat the runs with different -s values (e.g. from -s1 to -s100), to get significant results regarding the probability with which StirMark breaks your steganographic technique. -PSNR All options given above ignored in this case. Gives the PSNR of the input file compare to the file whose name is provided after the PSNR option: 2 N max X i i PSNR = 10 log_10 ------------------ N 2 Sum (X - X') i=1 i i -T All options given above ignored in this case. The tests implemented by default into StirMark include: - rotation by a small angle and cropping: -2, -1, -0.75, -0.5, -0.25, 0.25, 0.5, 0.75, 1 and 2 degrees. This simulates in part what is obtain when scanning an image: it is impossible to align it perfectly. - rotation by a small angle followed by cropping and rescaling to keep the original size of the image. - rotation by large angle: 5, 10, 15, 30, 45, and 90 degrees. - scaling by factors 0.5, 0.75, 0.9, 1.1, 1.5 and 2. - centered cropping: 1%, 2%, 5%, 10%, 15%, 20%, 25%, 50% and 75% - median filtering - Gaussian filtering 1 2 1 2 4 2 1 2 1 - sharpening 0 -1 0 -1 5 -1 0 -1 0 - horizontal flip - Frequency mode Laplacian removal attack - Colour quantisation (similar to GIF compression) - JPEG compression with quality factors: 90, 80, 70, 60, 50, 40, 35, 30, 25, 20, 15 and 10. Most existing algorithms break down towards 50, but certain images (e.g., baboon) still look fine at 30. Most pictures have a very poor quality with factor 10 but some algorithm survive up to this point. - Symmetric and asymmetric shearing (X and/or Y direction): (0, 1), (0, 5) (1, 0), (5, 0), (1, 1) and (5, 5). The first component is the shift in X direction (% of width) and the second is the shift in the Y direction (% of height). - Symmetric and asymmetric line and column removal: (1, 1), (1, 5), (5, 1), (5, 17), (17, 5). First component is the number of columns removed and the second component the number or rows. - General linear geometric transformation: |x'| |a b| |x| |y'| = |c d| |y| with the following parameters: (a, b, c, d) = (1.010, 0.013, 0.009, 1.011) (1.007, 0.010, 0.010, 1.012) (1.013, 0.008, 0.011, 1.008) - `StirMark' random geometric distortions Output files are in JPEG and PPM formats, depending on the test. Other information about these tests, how to use them can be found on: - S To be use with -T option only. This option can be used to select a category of transformation (e.g. rotation only). Please use ``stirmark -?'' to get the list of category numbers available. These numbers are subject to change. -------- Examples -------- - stirmark -Tbase test.ppm will apply all the test describe above and create one output file per test with base name "base" and different suffixes and extensions depending on the transformation applied. - stirmark -S34 -Tbase test.ppm will only apply category 34 of the tests. See ``stirmark -?'' for the list of available codes. - stirmark -PSNRoriginal.ppm modified.jpg asks StirMark to compute the PSNR of modified relatively to the original. - stirmark -NOJPEG -PS+sample.param sample.ppm sample_S+.ppm stirmark -NOJPEG -PL+sample.param sample.ppm sample_S+L+.ppm The first command applies geometrical distortions and save the parameters into sample.param. The second command loads these parameters and use them to apply the same geometric distortions. At the end sample_S+L+.ppm and sample_S+.ppm will be exactly the same. - stirmark -i0 -o0 -d0 test.ppm output.jpg Does not apply any geometric distortions. Just JPEG compression. -------------------------- About StirMark distortions -------------------------- Although many watermarking systems survive basic manipulations -- that is, manipulations that can be done easily with standard tools, such as rotation, shearing, resampling, resizing and lossy compression -- they do not cope with combinations of them and especially minor random geometric distortions. StirMark is a testing tool for watermarking systems that applies such distortions to images. The image is slightly stretched, sheared, shifted and/or rotated by an unnoticeable random amount. If A, B, C and D are the corners of the image, a point M of the said image can be expressed as M = a(b A + (1 - b)D) + (1 - a)(b B + (1 - b)C) where 0 <= a, b <= 1 are the coordinates of M relatively to the corners. The distortion is done by moving the corners by a small random amount in both directions. The new coordinates of M are given by the previous formula, keeping (a, b) constant. More distortions -- still unnoticeable -- can be applied to a picture. A global `bending' is applied to the image: in addition to the general bi-linear property explained previously a slight deviation is applied to each pixel, which is greatest at the center of the picture and almost null at the borders. On top of this a higher frequency displacement of the form l.sin(Ox.x)sin(Oy.y) + n(x,y) -- where n is a random number -- is added. All these distortions combined together with a mild JPEG compression do not remove the watermark per se. They prevent the detector from finding it. This suggests that the real problem is not so much inserting the marks as recognizing them afterwards. StirMark introduces a practically unnoticeable quality loss in the image if it is applied only once. However after a few iterated applications of the algorithm, the image degradation becomes soon very noticeable. This makes StirMark (especially the -d option) also a model for algorithms that can be applied to artificially degrade images on replay to get rid of the lossless-copying property of digital multimedia objects. --- Suggestions for improving this software are always welcome. We are also interested in robustness test results obtained by processing with StirMark images that were marked with various available watermarking tools. Surprisingly, first experiences with various watermarking tools show that the watermarks of *all* tools we have been able to locate until today can be removed or made undetectable by StirMark. For more information, please see: Still more information is provided in: Fabien A. P. Petitcolas, Ross J. Anderson and Markus G. Kuhn. Attacks on copyright marking systems. In David Aucsmith, Ed., second workshop on information hiding, in vol. 1525 of Lecture Notes in Computer Science Portland, Oregon, USA, 14--17 April, 1998, pp. 218--238. ISBN 3-540-65386-4. -------------------------- Copyright and legal notice -------------------------- StirMark is freely available under the GNU General Public Licence in full portable ISO C source code from Users should read the "Copyright" file provided in this package for copyright information about code and libraries used in StirMark. If you use StirMark for your research, please cite: Fabien A. P. Petitcolas, Ross J. Anderson, Markus G. Kuhn. Attacks on copyright marking systems, in David Aucsmith (Ed), Information Hiding, Second International Workshop, IH'98, Portland, Oregon, USA, April 15--17, 1998, Proceedings, LNCS 1525, Springer-Verlag, ISBN 3-540-65386-4, pp. 219--239. and Martin Kutter and Fabien A. P. Petitcolas. A fair benchmark for image watermarking systems, To in E. Delp et al. (Eds), in vol. 3657, proceedings of Electronic Imaging '99, Security and Watermarking of Multimedia Contents, San Jose, CA, USA, 25--27 January 1999. The International Society for Optical Engineering. To appear. Using StirMark for any other purpose than research or evaluation of copyright marking systems is prohibited, at least in Europe: "Member States shall provide adequate legal protection against the circumvention without authority of any effective technological measures designed to protect any copyrights or any rights related to copyright as provided by law or the sui generis right provided for in chapter III of European Parliament and Council Directive 96/9/EC". --------------- Acknowledgments --------------- - Peter Meerwald for the `man' page. - Martin Kutter , �cole Polytechnique F�d�rale de Lausanne, Switzerland, for providing some sample images for version 2.2. - This software is based in part on the work of the Independent JPEG Group - Alan Day for pointing out bugs into version 2.2, build 24. - Frank Hartung , Friedrich-Alexander- Universit�t Erlangen-N�rnberg, Germany, who pointed out a bug into image.c - Neil Dodgson , University of Cambridge, for providing the new resampling algorithms. - Richard Barnett , University of Essex, for the code of his Laplacian Removal attack. - Gabriela O. Csurka (Universit� de Gen�ve), Thomas Fiebig (Technische Universit�t Darmstadt), and Shelby Pereira for their valuable comments. Fabien. Cambridge, 6 April 1999. -- Fabien A. P. Petitcolas, The Computer Laboratory, University of Cambridge, New Museum Site, CB2 3QG, UK. $Header: /StirMark/README.txt 20 7/04/99 12:02 Fapp2 $