Currently, music scores are increasingly scanned, stored, and distributed in digital form. Although such music scores can be regarded as images, image watermarking techniques developed for greyscale or colour images can not be applied for watermarking music scores. The crisp and discrete graphics contain only little noise that could be used by an image watermarking scheme to hide information without visibly affecting the image quality (and without being easily removable). At the same time, artists are particularly critical about the quality of the scores they use.

Therefore we have developed different methods for this application, which are targeted to this particular problem. Rather than embedding information in arbitrary places, our methods hide information in or near the representation of syntactic elements in a music score. We developed three different methods: binary, symbolic, and staff line music score watermarking.

Binary Music Score Watermarking

This technique is based on a watermarking technique that we originally applied to binary images. For embedding information, the ratio of black and white pixels in certain image regions are modified. In contrast to binary images of natural objects, music scores do not contain irregular structures like textures. Therefore the modified pixels are located at edges to reduce visible artifacts. A further improvement of this method is achieved by a pre-segmentation of the input sheet music in musical symbols and objects like staff lines and slurs: Only the horizontal objects are used for embedding the watermark which increases the quality of the watermarked scores drastically.

	Original
	With watermark
	Watermark after segmentation

Symbolic Music Score Watermarking

In contrast to the above described method the symbolic method considers the content of the music scores. The representation of each musical symbol has some properties which can be modified. For example, the horizontal position or the width of the note heads can be changed by a certain amount. These degrees of freedoms are used to embed information in music scores. The advantage of this method is its increased quality and its increased robustness. It is related to text watermarking where for example similarities of different fonts are exploited for embedding information. Currently, we are working on reducing the artifacts (by appling image warping techniques) and on increasing the capacity of this method.

The examples given below show an original (unwatermarked) score and a watermarked score in which the horizontal distances of musical symbols are modified.

	Original
	With watermark

Staff Line Music Score Watermarking

The major draw back of the previous described method is the capacity which depends on the content of the music score which should be watermarked: The more symbols are available in the music score the more information can be embedded in it. To overbear this obstacle this method uses the staff lines for embedding information. This is done by modifications of the vertical distances between the staff lines and applying a slight bending. The robustness of this method is increased in comparision with the binary music score watermarking technique and the capacity is independent from the musical symbols available on a music score.

The examples given below show the original music score, the score watermarked by modifying the vertical distances of staff lines, and the score watermarked by bending the staff lines.

	Original
	With watermark (vertical distances modified)
	With watermark (staff lines bent)

Summary

All techniques described above can be used for watermarking music scores. They have varying features with regard to their capacity, imperceptibility, and robustness of watermarks. An advantage of these techniques is that they can be composed to achieve further improvements.