Digital watermarking

A digital watermark is a technical marker in carrier media ( works ) as image, video or audio files or texts, the plant-related and usually is non- perceptible. Unlike metadata digital watermark using steganographic methods are intertwined directly with the content to be marked. Digital watermarks are therefore seen as a commercial application of steganography. Example, they are used to embed copyright holders information on copyright infringement prosecution in a factory. In various research and development projects, which initially were limited to the area that procedures have been developed mainly for commercially relevant content such as audio and video data and 3D models. There are also methods, eg for marking databases, spatial data, music scores, still images and text.

Depending on the configuration through digital watermarking and digital -to-analog -to-digital conversions, ie the carrier medium does not necessarily have to be digital. A prominent example of such a watermark is found on Euro banknotes. Another example is digitally projected movies whose filmed as a digital watermark applied venue versions can be removed.

With digital watermark example, the authenticity of a file can be detected and their traceability be ensured. A carrier may contain several different watermarks at the same time. Unlike traditional watermarking digital watermarking are not directly perceptible by humans, but are intended only about a prescribed, also digital method to be detected and read.

Demarcation for steganography

In contrast to steganography is not the imperceptibility and secrecy of the Steganographieeinsatzes in the first place, but the robustness against attacks of digital watermarks in the rule. This can in extreme cases lead to noticeable changes of the carrier medium. Robust digital watermarks are usually designed so that the elimination of the digital watermark degrades the carrier into the Useless.

Criteria of watermark

Watermarks and their algorithms are classified according to different criteria, which are discussed in more detail in this section. You to understand is necessary with regard to the applicability in the present case.

The criteria are:

  • Blindness to the carrier
  • Perceptibility
  • Public
  • Fragility / robustness
  • Embedding space

Blindness to the carrier

Wherein the digital watermark is referred to as blind to the carrier if it is not needed to read the watermark. Accordingly, a method is not blind when the original signal is required when reading.

Normally, non-blind schemes are more robust than blind, because the differences between the watermark -bearing signal and the original can be found trivial. In practice, however, mainly meets Blind watermarking algorithms, since the lack of need to have access to the original data, makes the reading more flexible.


There are perceptible and imperceptible watermark. Noticeable watermarks are as logos or copyright notices that are subsequently introduced into images in order to identify the author. Imperceptible watermark, however, aim not to change the perceived impression of a work. The imperceptibility, was called by many scientists who conduct research in the field as a necessary criterion so that a watermark is a digital watermark.


Watermark algorithms can be distinguished in terms of their public availability or use of their range. Private algorithms can be detected only by appropriately authorized users. In these algorithms, priority is placed on ensuring that unauthorized users can not read the private watermark. In contrast to these algorithms private public watermark for each can be read.

In general, private watermarks are more robust than public, because it is easier for an attacker in public watermark by the knowledge about the watermark, make noise or distances.

Fragility / robustness

One of the main criteria by which to distinguish watermark, is the fragility ( fragility ) or - inversely thereto - the robustness. Robust digital watermarks survive depending on the method processing steps such as Crop, Enlarge / Reduce, digital -to-analog -to-digital conversion, etc. This kind of digital watermark is widespread due to the use field of copyright infringement prosecution. In contrast, fragile method for the detection of integrity and authenticity can be used. The methods fall into this category, can be again broken down into global approaches, which allow only binary statements, and local approaches, can be shown by the manipulated regions of the support, since there the digital watermark is destroyed.

There are hybrid forms of fragility, called semifragile watermark. Thus, methods exist that behave robust against certain steps, towards other but fragile. For example, the lossy storage can be tolerated, but not a processing such as cutting.

Embedding space

For the embedding of a watermark signal can be modified both as such as well as in a transformed state. The former is called a modification in the spatial domain or in audio signals as a period of time. Transformed signals are changed in the frequency domain. Examples of transformations are Fourier transform, discrete cosine transform or wavelet transform.

The embedding in the frequency domain generally leads to an increased robustness, as for example an amplitude normalization in the spatial domain has no significant effect on the transformed signal.


The type of the embedded message is determined by the application. This may, in the case of the proof of authorship may be information on the copyright owner, for example. In the general case metadata is embedded. ( engl. meta data labeling ).

Due to practical limitations on message length (capacity) but all the metadata can not be embedded directly. Accordingly, a much shorter link information is embedded to the detailed data on the medium (see, primary key).

Examples of specific applications include:

However, the designations for various applications are not always clear. Thus, the term fingerprinting is used instead of the term transaction watermark (similar to the idea that the customer leaves its fingerprint on its contents ). Also the term fingerprinting is used as well as perceptual hashing method that calculate a characteristic digital fingerprint for the digital content for the detection of a content based on their intrinsic characteristics.

Both classes of methods - digital watermarking and digital fingerprinting - belong to the class of passive protection mechanisms. In contrast to active protection mechanisms, such as encryption, these mechanisms do not prevent unauthorized access to content.

They are therefore considered by various groups as a better alternative to DRM. In particular, by their property of robustness against format conversions, and various operations they can cover a wide range of applications.

A special class are the reversible watermarking techniques. Here, the embedded watermark can be removed and the original message is recovered. You apply the recovery information additionally to the newly introduced ( watermark ) data in the watermark. Applications of this Wasserzeichenart found, for example in medical image processing.


Watermarking techniques have different properties:

  • The visibility means the influence of the quality of the selected content. To evaluate the visibility in comparison to the original differential perception threshold is commonly used
  • The durability, as described above, and
  • The capacitance is determined by the amount of information that can be introduced into a medium. It ranges from binary watermark ( labeled or unlabeled ) (depending on disk size and method ) to watermark with several hundred bytes capacity.

These characteristics are dependent on each other. One can illustrate this with a triangle at the corners are these three properties. If two parameters are selected, the result is the third parameter.

Other characteristics that distinguish different methods are

  • The (non) detectability, which indicates how easily a watermark by analyzing the processed data stream can be detected (eg through a spectrogram or web analytic methods), and has a strong correlation with the
  • Attack security (English Security ), which defines the cost to be operated at the wanton removal of marking and depends on the reversibility of the embedding algorithm, and
  • Blindness, which expresses whether to read out the information of the watermark, the original data set is required, or whether the embedded data can be obtained directly from the marked data stream.

The significance of the individual features will depend on the particular application. Usually the visibility is the most important criterion followed by the robustness and capacity.

In concrete terms, the dependence described above, that visibility can not be chosen independently of the robustness: the less an embedded message is to be perceived, the lower the potential changes in the content without coming on the perception threshold. At the same time, these changes can also be more easily removed.

There are also other features depending on the particular application scenario, such as watermarking minimum segment (WMS, smallest subdivision of a content, a complete watermark message contains ) or the real-time capability of the algorithm.