Classification of Watermarks

Fragile

The fragile watermark is one of the watermarking methods for authentication that has a low robustness toward modifications where even small changes of the content will destroy embedded information, showing that there has been an attempt of attack. A good example is medial records.

A fragile watermark is supposed to break. Think of it as a digital version of the foil put under the cap of a medicine bottle. It is pretty hard to get by that without it looking like it has been tampered with. Watermarks are a digital version of this idea.

Robust

A robust watermark is almost exactly the opposite of a fragile watermark. A robust watermark can be either visible or invisible, depending on purpose. Robust watermarks are very difficult to remove or damage. Following are some things to which a robust watermark can be resistant:

• Robustness to cropping: The watermark is embedded in so many places throughout the image that removing part of it will not remove the watermark.

• Robustness to scaling and rotation: This gets a bit more complicated, and without getting into the math, the simplest way to put it is the watermark algorithm has scale and rotation parameters built in that allow it to apply these "variances" when examining a watermark. These calibrated variances allow the watermark to withstand a measure of scaling and rotation without making them unrecognizable to the algorithm.

• Robustness to translations: When an image is translated, an effect called zero padding is introduced, which consists of appending zeros to a signal. This zero padding occurs, for example, when an image is scanned in from a hard-copy photo. Generally, this type of condition can be overcome by the watermarking algorithm because it is able to look for the watermark in the right place on the image and reconstruct the watermark with only about 1/3 of the watermark block.

• Robustness to compression: When compression takes place, the compression regularly happens on specific frequencies of the image (typically low frequencies). With this knowledge the watermark can be embedded in higher frequencies, thus maintaining its integrity when the lower frequencies are stripped away during compression, causing a reduction in size and quality such as during JPEG compression.

• Robustness to other attacks: Namely, a StirMark attack, which simulates the distortions an image would go through if it were scanned, printed, and photocopied:

  1. Unobtrusive: While a watermark is supposed to be visible, it should be so in a manner that does not interfere with what is being protected.

  2. Robust: The watermark should be difficult or impossible to remove, and should be immune to:

     1. Common signal processing: Analog-to-digital and digital-to-analog conversion, resampling, and changes to contrast, color, etc.

     2. Common geometric distortions: Rotation, cropping, scaling, etc.

     3. Subterfuge attacks: Combining multiple data sets to destroy a watermark.

  3. Universal: The watermarking algorithm should apply to all media, image, audio, and video, which also adds assistance when watermarking multimedia.

  4. Unambiguous: Retrieval of a watermark should identify the owner without question.