Department of Optimization of Numerical Methods

Computer Steganography and watermarking

The word «steganography» in a translation from Greek means secret writing (steganos – secret, a secret; graphy – writing). The history of steganography counts a millennium. Concealment the fact of existence confidential messages has always been considered expedient for its protection, and existence various technical, chemical, physical and psychological methods of such concealment provided possibility its realization. Steganography has been actively researched and developed with the advent of computer technologies.

Computer steganography develops in the several directions having both similar lines, and the certain differences caused by features of practical use. Among steganosystems allocate systems of the hidden data transmission, digital watermarks, identification numbers ("fingerprints") and captions.

Digital watermarking is the process of embedding some information (that is called watermark or mark) into a digital multimedia content such way that makes this mark imperceptible and irremovable after some modifications of the carrier. Often, the host signal that carries the watermark is also called a cover object.  Digital watermarking systems are usable for a wide range of practical applications, such as noise-robust authentication of audio and visual data (in particular, for the integrity control of CCTV- or telephone conversations recordings), authentication of the data owner (copyright protection), authentication of the data source,  broadcasts monitoring, copying control etc.

In general, a digital watermarking system can be considered as a set of original cover objects X, watermarks W, keys K, marked cover objects Y; and watermark embedding and extracting transformations which are denoted as E and D accordingly.

Figure 1: General model of the watermark life-cycle with embedding, attacking, and detecting/decoding transformations

Watermarking systems have to satisfy the following requirements: 

  • Imperceptibility.
  • Robustness.
  • Security.
  • Reliability (or detection rate of the watermark).
  • Capacity.
  • Speed.

Audio watermarking method robust against lossy compression

Here is an example of the audio watermarking method, which implements the scheme with the watermark decoder, and in which watermark can be restored after the lossy compression attacks. The method is based on the fact that the watermark is placed in a signal form which is more stable and predictable parameter of the attacked cover than the individual samples values.

An initial audio signal is divided into equal-length blocks. The length is a question of compromise between high resolution in frequency domain during subsequent spectral analysis and computational complexity of the algorithm. One bit of additional information is embedded in each block.

On the first step of embedding procedure, coefficients of frequency subbands of the signal block, which will serve as an immediate carrier of digital watermark bit, are determined using FWT or PWT. On the next step subband-carrier spreads by a secret key that is pseudo-random uniformly distributed sequence of 1 and -1. The method of slow spread spectrum  is used on this step.

The key determines the location of current watermark bit into the current frequency subband, which will be the carrier of this bit. The watermark extraction using another key than the one used for embedding, will lead to the extraction of bits out of the wrong locations, i.e. extract random values. The key is independent of the signal, and its length is equal to the length of subband-carrier.

Then using the Fourier transform, a transition to the frequency representation of the spread subband by key sequence makes; and amplitude spectrum coefficients are considered hereinafter as coefficients for embedding.

The location of current bit into current signal block is determined by the location of maximum coefficient of the amplitude spectrum. This location (with sufficient frequency resolution) is invariant to the lossy compression. Then, the method provides removing (zeroing) of the three coefficients on the left from the maximum if zero-bit value is embedded, or three coefficients on the right from the maximum for one-bit value. Thus, the digital watermark is encoded as relative difference between the samples.

The extraction procedure is identical to embedding procedure until the determination of maximum amplitude location in the Fourier domain of the subband-carrier. Further, if the sum of three coefficients on the left of the maximum is bigger than the sum of three coefficients on the right from it, then a one-bit value is extracted from the current block, and if vice versa - zero-bit value.

This method was implemented using Matlab package. Two-level FWT based on Daubechies wavelet of order 10 was used to determine the subband-carrier. Audiocodec that was used during robustness testing – LameXP.

Robustness was estimated as the relation between number of correctly extracted bits and number of embedded (Ratio of Correct Bits Recovered). Results of watermark robustness testing against compression according to standards MPEG-1 Layer 3, Ogg Vorbis are given in Fig. 2 and Fig. 3.

Figure 2: Results of watermark robustness testing against the MPEG-1 Layer 3 compression attack

Figure 3:  Results of watermark robustness testing against the Ogg Vorbis compression attack

It is possible to improve ROCBR by using redundant digital watermarks.

In this work, also other methods of introducing digital watermarks to audio and images that allow for retaining the marks of a signal after apllying to it algorithms of compression with losses and other modifications are proposed. . Possibilities of using subband coders and transform coders in building steganography systems are shown, as well as ways of introducing a key of access legality in steganosystems with watermark, increasing system reliability, and coding  watermark bits. The  methods can be applied to resolve  issues such as identification, authentication, protection of copyrights, and controlling use of audio data and image.

The self-synchronized speech watermarking  method and methods for protection of intellectual property on paper carriers also was created. Natural indignations in the channel of printing and scanning are parsed, their influence on an amplitude spectrum of the map is studied.

  • Koshkina N.V.  Methods of synchronizing digital watermarks // Cybernetics and Systems Analysis,  Volume 44, Issue 1, 2008, P. 145 – 152.  (
  • Koshkina N.V.  Method of Audio Signals Watermarking Based on the Wavelet and Fourier Transforms // Journal of automation and information sciences, 2010, № 42, P. 71–80.
  • Koshkina N., Zadiraka V., Smolarz A., Komada P. Methods of watermarking //Current problems in information and computational technologies /edited by W. Wojcik, J. Sikora. – Lublin: Politechnika Lubelska, 2012. – Vol. 1. – P.61-98.
  • Koshkina N.V. Robust Self-synchronised Digital Speech Watermarking System // Journal of automation and information sciences, 2012, № 44, P. 72–82.
  • Koshkina N.V. Detection of Hidden Messages Embedded in Audio Signals by Hide4PGP // Journal of automation and information sciences, 2013, № 45, P. 75–81.
  • Koshkina N.V. Investigation of the Applicability of the Co-occurrence Matrix for Detecting Steganoaudiosignals // Journal of automation and information sciences, 2013, № 46, P. 63–72.
  • Zadiraka V., Koshkina N. Spectral methods of computer steganography problem decision //Methods of effective protection of information flows /edited by V. Zadiraka, Y. Nykolaichuk. – Ternopil: Terno-graf, 2014. – P. 96-120.