Learning to remove staff lines from music score images

Igor S. Montagner, Roberto Hirata Jr. and Nina S. T. Hirata

Abstract

The methods for removal of staff lines rely on characteristics specific to musical documents and they are usually not robust to some types of imperfections in the images. To overcome this limitation, we propose the use of binary morphological operator learning, a technique that estimates a local operator from a set of example images. Experimental results in both synthetic and real images show that our approach can adapt to different types of deformations and achieves similar or better performance than existing methods in most of the test scenarios.

Software

We used TRIOSlib to train the operators used in this work. Some guides on how to use our software are available at our tutorials section, but feel free to contact us if encounter any difficulties.

Dataset 1 - synthetic images

The synthetic images and the code for all deformations and methods were obtained from the Music Staves package. Dalitz et al [1] describe 8 types of deformations that can be applied to . See in Figure 1 an example of the effect of each deformation in a sample image.

(a) Original image.

(b) Curvature.

(c) Thickness.

(d) White speckles.

(e) Interruptions.

(f) Typeset.

(g) Kanungo

(h) Y-Variation.

(i) Rotation.
Figure 1. Deformations used to test the robustness of staff removal methods.
The parameters for the deformations were taken from [1] and are shown in Table 1.
DeformationParametersValues
Curvature Amplitude/Width 0:0.02:0.1
Thickness Minimum height
Maximum height
Inertia
2
10
0.5:0.1:0.9
White speckles Smoothing factor
Random walk length
Speckle frequency
2
10
0.0:0.01:0.1
Interruptions Frequency
Width
0:0.01:0.05
0:1:9
Typeset Maximum gap
Maximum shift
0:1:11
0:1:9
Kanungo Smoothing Factor
Random walk length
a, b
a0, b0
2
10
[0.5,1]
0.25:0.25:1.5
Y-Variation Inertia
Maximum deviation
0.8:0.05:0.95
0:1:5
Rotation Angle -3:1:3
Table 1. Parameter values possible for each deformation given as min:step:max.
Finally, the results for each deformation are presented in the links below.

Dataset 2 - handwritten images

The set of handwritten images were obtained from the "ICDAR/GREC 2013 Music Scores Competition on Staff Removal"[2] (available online).

References

[1] C. Dalitz, M. Droettboom, B. Pranzas, and I. Fujinaga. A comparative study of staff removal algorithms. IEEE Trans. Pattern Anal. Mach. Intell., 30(5):753-766, 2008

[2] Visaniy, M.; Kieu, V.C.; Fornes, A.; Journet, N., "ICDAR 2013 Music Scores Competition: Staff Removal," Document Analysis and Recognition (ICDAR), 2013 12th International Conference on , vol., no., pp.1407,1411, 25-28 Aug. 2013