DAG

The CVC-MUSCIMA Database

Overview

The CVC-MUSCIMA database contains handwritten music score images, which has been specially designed for writer identification and staff removal tasks.

The database contains 1,000 music sheets written by 50 different musicians. All al them are adult musicians, in order to ensure that they have their own characteristic handwriting style. Each writer has transcribed the same 20 music pages, using the same pen and the same kind of music paper (with printed staff lines). The set of the 20 selected music sheets contains music scores for solo instruments and music scores for choir and orchestra.

Furthermore, for the staff removal task, each music page has been distorted using different transformation techniques, which, together with the originals, yield a grand total of 12,000 images.

The database and ground-truth are fully described in [1].

Characteristics


The original 1,000 music scores have been scanned at 300 dpi and 24 bpp. Later, they have been converted to 8 bit gray scale.
The staff lines were initially removed using color cues, and manually checked for correcting errors.
Finally, the music scores have been distorted for staff removal purposes and converted to binary images.

Two different ground-truthed images have been created, depending whether the aim is writer identification or staff removal.

Writer Identification

Each one of the original 1,000 images is labelled with its writer identification code, and presented in three flavours: grey scale image (at 256 grey levels), binary image, and staff-less binary image. Thus, the provided staff-less images will allow the research on writer identification techniques with independency of the staff removal technique applied.






Staff Removal

To test the robustness of different staff removal algorithms, we have applied the following distortion models proposed by Dalitz et. al [2].

Degradation with Kanungo noise Rotation Curvature
Staffline interruption Typeset emulation Staffline y-variation
Staffline thickness ratio Staffline thickness variation White speckles


We have applied the nine distortions, where two of them (Staffline y-variation and Staffline thickness variation) have been applied twice with different parameters. In summary, we have obtained 11,000 distorted images. The parameters are described here.

Thus, for each one of the 12,000 binary images (the 1,000 original images plus the 11,000 distorted images), we present the images with music symbols and staff lines, images with only the staff lines, and images without the staff lines (staff-less images).




Partitions


For facilitating the comparison among different approaches, we devised two sets of ten partitions:


Download

The set of images in PNG format are available to download.

The partitions are available in TXT and MAT format (for Matlab users):

The staff distortion code is available for letting the users generate staff distortions with different parameters:

If you are interested in the set of raw images in BMP format (2.7Gb), please contact us.



References

[1] Alicia Fornés, Anjan Dutta, Albert Gordo, Josep Lladós. CVC-MUSCIMA: A Ground-truth of Handwritten Music Score Images for Writer Identification and Staff Removal. International Journal on Document Analysis and Recognition, Volume 15, Issue 3, pp 243-251, 2012. (DOI: 10.1007/s10032-011-0168-2).


[2] Christoph Dalitz, Michael Droettboom, Bastian Pranzas and Ichiro Fujinaga. A Comparative Study of Staff Removal Algorithms. IEEE Transactions on Pattern Analysis and Machine Intelligence 30(5), pp.753-766. 2008. (DOI: 10.1109/TPAMI.2007.70749) (the authors have a free archived version here).



Contact

 If you have any questions or suggestions, please contact Alicia Fornes: afornes@cvc.uab.es.



Terms of Use


This dataset is for non-commercial research purpose only.

License:
Creative Commons License
This dataset is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Reference:
If you publish material based on this database, we request you to include a reference to paper [1].

Contributors:
Alicia Fornés, Anjan Dutta, Albert Gordo, Josep Lladós.