STAFF LINE DETECTION AND REMOVAL WITH STABLE PATHS

Artur Capela, Ana Rebelo

INESC Porto, Campus da FEUP, Rua Dr. Roberto Frias 378, 4200-465 Porto, Portugal

Jaime S. Cardoso

INESC Porto, Faculdade de Engenharia, Universidade do Porto, Portugal

Carlos Guedes

INESC Porto, Escola Superior de M

usica e Artes do Espect

aculo, Portugal

Keywords:

Music, optical character recognition, document image processing, image analysis.

Abstract:

Many music works produced in the past are currently available only as original manuscripts or as photocopies.

Preserving them entails their digitalization and consequent accessibility in a machine-readable format, which

encourages browsing, retrieval, search and analysis while providing a generalized access to the digital material.

Carrying this task manually is very time consuming and error prone. While optical music recognition (OMR)

systems usually perform well on printed scores, the processing of handwritten music by computers remains

below the expectations. One of the fundamental stages to carry out this task is the detection and subsequent

removal of staff lines. In this paper we integrate a general-purpose, knowledge-free method for the automatic

detection of staff lines based on stable paths, into a recently developed staff line removal toolkit. Lines

affected by curvature, discontinuities, and inclination are robustly detected. We have also developed a staff

removal algorithm adapting an existing line removal approach to use the stable path algorithm at the detection

stage. Experimental results show that the proposed technique outperforms well-established algorithms. The

developed algorithm will now be integrated in a web based system providing seamless access to browsing,

retrieval, search and analysis of submitted scores.

1 INTRODUCTION

The Universal Declaration on Cultural Diversity

adopted by the General Conference of UNESCO on

2001 asserts that cultural diversity is as necessary for

humankind as biodiversity is for nature, and that poli-

cies to promote and protect cultural diversity thus are

an integral part of sustainable development. Being

music a pivotal part of our cultural heritage, its preser-

vation, in all of its forms, must be pursued. Fre-

quently, the preservation of many music works en-

tails their digitalization and consequent accessibility

in a format that encourages browsing, analysis and re-

trieval.

There is a vast amount of invaluable paper-based

heritage, including printed and handwritten music

scores, which are deteriorating over time due to natu-

ral decaying of paper and chemical reaction (i.e., be-

tween written ink and paper). Various efforts have

been focused on this issue in order to preserve the

record of our heritage. Digitisation has been com-

monly used as a possible tool for preservation. Al-

though a digital copy may not conserve the original

document, it can preserve the most important part:

its data. It has also the advantages of easy dupli-

cations, distribution, and digital processing. Never-

theless, the output of the digitalization process is not

amenable for further analysis or semantic search op-

erations. Thus, an Optical Music Recognition (OMR)

process is needed. However, the manual process re-

quired to recognize handwritten musical symbols in

scores and to put them in relationship with the spine

structure of the score is very time consuming. This

justiﬁes the research around reliable automatic OMR

algorithms as current solutions are still below the ex-

pectations.

As a concrete example, Portugal has a notorious

lack in music publishing from virtually all eras of

its musical history. However, whereas most of the

known original music manuscripts before the twen-

263

Capela A., Rebelo A., S. Cardoso J. and Guedes C. (2008).

STAFF LINE DETECTION AND REMOVAL WITH STABLE PATHS.

In Proceedings of the International Conference on Signal Processing and Multimedia Applications, pages 263-270

DOI: 10.5220/0001937802630270

 SciTePress

tieth century are kept at the National Library Archive

in Lisbon, there is virtually no national repository for

the Portuguese music from the twentieth century. Al-

though there are recent efforts in order to catalogue

and preserve in digital form the Portuguese music

from the late twentieth century—notably the Music

Information Center (MIC, 2008) and the section on

musical heritage from the Institute of the Arts web-

site (IOA, 2008)—most of the music pre-dating com-

puter notation software was never published and still

exists as manuscripts or photocopies spread out all

over the country in inconspicuous places. The risk

of irreversibly losing this rich cultural heritage is thus

a reality.

1.1 OMR System

The project “Optical recognition system for handwrit-

ten music scores” initiated in 2007 by INESC Porto

and ESMAE is the point of departure for creating

a web-based system of music manuscripts of Por-

tuguese composers from the twentieth century. This

database will provide generalized access to a wide

corpus of unpublished handwritten music encoded in

MusicXML, which can be accessed remotely via the

Internet. The database will not only centralize as

much information as possible but will also serve to

preserve this corpus in a way that is easily accessi-

ble for browsing, analysis, and ultimately, for per-

forming this repertoire, therefore helping to keep the

Portuguese music alive (Capela et al., 2008). Al-

though the aim of this project is Portuguese music,

it is equally valid for all printed and handwritten mu-

sic scores that need to be preserved from all around

the world.

The ambitious goal of providing generalized ac-

cess to handwritten scores that have never been pub-

lished has been severely hampered by the actual state-

of-the-art of handwritten music recognition. There

are currently various commercial OMR software so-

lutions (Capella software

, SharpEye Music Reader

OMeR

) and a few open source solutions (AOMR2

OpenOMR

, Audiveris

), but they are all ofﬂine stan-

dalone applications. The existing online archives of

music scores (Lester Levi Collection, 2008; Classical

http://www.capella-software.com/capscan.

htm

http://www.visiv.co.uk

http://www.myriad-online.com/en/products/

omer.htm

http://www.bzzt.net/

arnouten/wiki/index.

php/Gamera#AOMR2:_omr_toolkit

http://sourceforge.net/projects/openomr

http://audiveris.dev.java.net

Sheet Music Collection, 2008; Mutopia Collection,

2008) usually provide them in inadequate formats—

usually only as the scanned score image—for retrieval

or automatic analysis. These online archives are mere

standard websites, without facilities for optical recog-

nition, editing and searching through the scores mu-

sical content. The creation of an OMR system, in-

tegrating optical recognition, storage, search, brows-

ing and downloading capabilities, while keeping the

scores in their original format along with their digital

counterpart, would therefore be extremely beneﬁcial.

An integrated score editor would be provided in order

to view and edit the submitted music scores.

In our previous work on this project we have

presented a complete OMR System solution—

OMRSYS (Capela et al., 2008)—comprising a

database driven web application with one or more

OMR applications integrated in the proposed system.

The proposed architecture successfully attends the

stated objectives. At the end of our project we plan

on developing a fully functional system according to

the speciﬁed architecture and integrating a complete

OMR package.

1.2 Detection and Removal of Staff

Lines

Staff line detection and removal are the ﬁrst funda-

mental stages on the OMR process, with subsequent

processes relying heavily on their performance. The

reasons for detecting and removing the staff lines lie

on the need to isolate the musical symbols for a more

efﬁcient and correct detection of each symbol present

on the score. Although their primary application is as

a preprocessing step in the recognition of music no-

tation, the line detection problem also occurs in dif-

ferent contexts (e.g., the recognition of bank transfer

forms).

The detection of staves is complicated due to a

variety of reasons. The handwritten staff lines are

rarely straight and horizontal, and are not parallel to

each other. For example, some staves may be tilted

one way or another on the same page or they may be

curved. These scores tend to be rather irregular and

determined by a person’s own writing style. More-

over, if we consider that most of these works are old,

the quality of the paper in which it is written might

have degraded throughout the years, making it a lot

harder to correctly identify its contents.

In (Cardoso et al., 2008a; Cardoso et al., 2008b)

we presented a new and robust staff line detection al-

gorithm based on a stable paths approach. The pro-

posed paradigm uses the image as a graph, where the

staff lines result as connected paths between the two

SIGMAP 2008 - International Conference on Signal Processing and Multimedia Applications

264

lateral margins of the image. A staff line can be con-

sidered as a connected path from the left side to the

right side of the music score. As staff lines are almost

the only extensive black objects on the music score,

the path to look for is the shortest path between the

two margins if paths (almost) entirely through black

pixels are favoured.

In this paper we present our recent work and re-

sults focusing on the implementation of the Stable

Paths algorithm as a C++ plugin for the MusicStaves

Toolkit (Dalitz et al., 2008; MusicStaves Toolkit,

2008) (based on the Gamera Framework (MacMillan

et al., 2002; Gamera Framework, 2008)), as well as

on the removal stage by using our detection algorithm

in the ﬁrst stage. We have also adapted a removal al-

gorithm based on the LineTrack Height approach pro-

posed in (Dalitz et al., 2008), which we will present in

section 3. In section 2 the Gamera and MusicStaves

Toolkit are presented, together with our C++ plugin.

In section 4, both our proposed detection and removal

algorithms are evaluated experimentally using a well-

known dataset of music scores. Finally, conclusions

are drawn and future work is outlined in section 5.

2 STABLE PATHS INTEGRATION

In this section we present the platform in which our

detection algorithm was integrated for testing and val-

idation. The platform is comprised by its core—the

Gamera Framework (MacMillan et al., 2002; Gamera

Framework, 2008)—and by a toolkit for the evalua-

tion of detection and removal algorithms—the Mu-

sicStaves Toolkit (Dalitz et al., 2008; MusicStaves

Toolkit, 2008). This toolkit is a set of Gamera plu-

gins aiming to support the development and test of

staff line detection and removal algorithms for music

scores, by extending the Gamera functionality.

After we describe each part constituting this plat-

form, we present the implementation of our staff line

detection algorithm as a C++ plugin on the Music-

Staves Toolkit. Finally, we present the integration of

our detection algorithm on some removal algorithms

in the MusicStaves toolkit. We have integrated our

algorithm in those removal algorithms where the de-

tection is processed separately from the removal op-

eration, replacing the Dalitz algorithm (Dalitz et al.,

2008) as the detection stage.

2.1 Gamera Framework

Gamera (MacMillan et al., 2002; Gamera Framework,

2008) is a portable and open source framework to cre-

ate structured documents analysis applications by do-

main experts. The name Gamera is an acronym to

“Generalized Algorithms and Methods for Enhance-

ment and Restoration of Archives”. It combines a pro-

gramming library with graphical tools for an interac-

tive training and development of recognition systems.

This framework tries to be a tool to create custom ap-

plications through the domain experts knowledge in-

stead of responding to various requirements with a

monolithic application. It aims at providing an ef-

ﬁcient test and reﬁnement development cycle. The

programming language in its basis is the high-level

language Python, although it has many extensions

written in C++ to carry out low-level image process-

ing. Nevertheless, due to its nature, Python turns the

code writing process more agile and facilitates the use

of scripting, which makes Gamera an interactive and

batchable framework. Besides the large number of ex-

tensions deployed with this framework, it is also pos-

sible to customize and extend with plugins or toolkits,

written in Python and C++.

The Gamera framework is modular and organized

in a series of horizontal layers, which can be seen at

Figure 1.

Figure 1: Gamera Architecture.

Gamera follows a modular plugin architecture. It

is made of modules (plugins), both written in Python

and C++, integrated in a high-level scripting environ-

ment. Each module executes a task on the recognition

process. The framework maintains a toolbox design

approach, i.e., a user has access to a large set of tools

for the optical recognition stages.

2.2 MusicStaves Toolkit

MusicStaves (Dalitz et al., 2008; MusicStaves

Toolkit, 2008) is a Gamera toolkit speciﬁc for the de-

velopment and test of staff line detection and removal

STAFF LINE DETECTION AND REMOVAL WITH STABLE PATHS

265

algorithms. This Phyton toolkit also integrates facili-

ties to create a test set of music scores and to evaluate

results with established metrics. As with Gamera, this

toolkit is portable and its source code is freely avail-

able. In order to use MusicStaves, it must be imported

to Gamera, either in the GUI or in a programmatic

manner.

MusicStaves is structured as seen in Figure 2. The

toolkit is composed by a set of main classes where the

staff line detection and removal algorithms are imple-

mented, and by a set of plugins in Python and C++.

Some plugins are used by the implemented algorithms

while others are tools to aid the algorithms testing. It

is an extendable toolkit in which one can integrate a

new staff line detection and/or removal algorithm. Its

plugin system follows the Gamera framework ratio-

nale and as such the plugins may be written in Python

and C++. In order to write new staff line detection

or removal algorithms the toolkit provides two inter-

faces: StaffFinder (detection) and MusicStaves (re-

moval).

Figure 2: MusicStaves Architecture.

A test set of 32 synthetic music scores is also pro-

vided by the authors of this toolkit. Moreover, the

toolkit allows applying a set of deformations (e.g., ro-

tation, curvature, typeset emulation, white speckles)

commonly found in the real world to these perfect

scores—see Figure 3. The purpose is to be able to

measure the performance of the removal algorithms

contained in MusicStaves by using three deﬁned er-

ror metrics (Dalitz et al., 2008): Pixel Level, Seg-

mentation Region Level and Stafﬂine Interruptions.

However, the same set may be used to evaluate staff

line detection algorithms alone by deﬁning adequate

error metrics (Cardoso et al., 2008a; Cardoso et al.,

2008b). We have restricted the range of the param-

eters controlling the intensity of the deformations to

values considered realistic.

Figure 3: Some examples of applied deformations: a) Cur-

vature; b) Rotation; c) White Speckles; d) Stafﬂine Inter-

ruptions; e) Stafﬂine y-variation; f) Typeset Emulation. See

(Dalitz et al., 2008) for details.

2.3 Integration

We have integrated our recently proposed algorithm

for staff line detection (Cardoso et al., 2008a; Car-

doso et al., 2008b) in MusicStaves as a C++ plugin.

The plugin encompasses the main StaffFinder class in

the toolkit root, the respective Python plugin in the

Python plugins folder and the algorithm implementa-

tion in C++ in the C++ plugins folder. In Figure 4

we present a diagram of the algorithm as integrated in

MusicStaves.

The algorithm processing starts with the call to the

method ﬁnd staves, which receives a binarized image

as input. That image is then passed to the function

in C++ by the plugin Python code. In the C++ im-

plementation, after the respective staff line detection

function is called, the image is converted to the format

used internally by our algorithm. After the whole de-

tection task is complete it returns the staff lines skele-

ton in the MusicStaves format. After receiving the

skeleton list on its Python class, it ﬁlls the structure

self.linelist with the obtained values.

Figure 4: Overall view.

Besides integrating our Stable Paths Approach al-

gorithm into the MusicStaves toolkit as a C++ plu-

gin, we have also integrated the algorithm with staff

SIGMAP 2008 - International Conference on Signal Processing and Multimedia Applications

266

removal algorithms in order to evaluate the improve-

ments over the original results. However, this is not

a standard integration as the toolkit does not provide

the means for this kind of integration. This integra-

tion was coded in the staff removal algorithms by

adding the possibility to choose the staff line detec-

tion algorithm through a parameter value. A diagram

illustrating this integration can be seen in Figure 5.

Some removal algorithms present in MusicStaves de-

tect the staff lines along with the removal process.

From those, we have used the one with best results

in general (Dalitz et al., 2008)—Skeleton—for com-

parison purposes.

Figure 5: Stable Paths integration with staff removal algo-

rithms.

3 STAFF REMOVAL

ALGORITHMS

In the process of recognizing music scores the staff re-

moval algorithm is processed after the staff line detec-

tion stage takes place. In our current work the removal

algorithm is based on the LineTrack Height algorithm

presented on (Randriamahefa et al., 1993). The goal

on this method is to track the staff lines positions ob-

tained by a detection algorithm and remove vertical

run sequences of black pixels that have a value lower

than a speciﬁed threshold, which was experimentally

set at 2*staffHeight.

As music scores may suffer from deformations,

the staff lines may have descontinuities, be curved or

inclined. These problems will inﬂuence the success

to achieve a correct detection of lines contained on

the score to recognize. However, due to the above-

mentioned problems, the positions of the staff lines

obtained by a staff line detection algorithm may pass

slightly above or under the real staff lines positions.

That way, if we are in presence of a white pixel

when the staff lines are tracked, we search vertically

for the closest black pixel. If that distance is lower

than a speciﬁed tolerance—experimentally chosen as

1+ceil(staffHeight/3.0)—we move the reference posi-

tion of the staff line to the position of the black pixel

found.

Algorithm 1 : Staff Removal Algorithm.

procedure STAFFLINEREMOVAL(IMAGE,STAV ES)

threshold = 2∗staffHeight;

tolerance = 1 + ceil(staffHeight/3.0);

IMAGE REMOV E = copy(IMAGE);

for nvalid = 0 to STAVES size do

Point2D staff = validStaves[nvalid];

for i = 0 to staff size do

col =staff[i].x;

re f Row =staff[i].y;

row = re f Row;

pel = valuePixel(IMAGE,IMAGE REMOV E)

decrement/increase the reference row until one

pixel different from white pixel (dist1/dist2) is found;

if dist1 ≤ max(1, min(dist2,tolerance)) then

re f Row− = dist1;

else

if dist2 ≤ max(1,min(dist1,tolerance))

then

re f Row+ = dist2;

else

continue;

end if

Count the number of decrements/increase on

the reference row until the black pixel changes to white

pixel (run);

if run ≥ threshold then

continue;

end if

remove the vertical black sequences on the IM-

AGE;

end for

end procedure

On (Dalitz et al., 2008) a new method is

presented—Skeleton—which uses the skeleton infor-

mation, but performs the staff removal on the original

image instead of the skeleton. The method relies on

the fact that symbols on the stafﬂines lead to junction

points or corner points in the skeleton.

4 RESULTS

Although the evaluation of new staff detection algo-

rithms may be done by visually inspecting the out-

put on a set of scores—as adopted on (Rebelo et al.,

2007)—, our current comparison is supported on

quantitative measures. The test set adopted for the

qualitative evaluation of the proposed method is the

one presented in (Dalitz et al., 2008) and already de-

scribed. It consists of ideal synthetic scores to which

a set of known deformations have been applied—see

STAFF LINE DETECTION AND REMOVAL WITH STABLE PATHS

267

(Dalitz et al., 2008) for more details. In total we have

generated 2688 deformed images originated from 32

perfect scores. In order to conveniently measure the

performance of staff line removal algorithms we have

adopted two error metrics from (Dalitz et al., 2008):

Pixel Level and Segmentation Region Level.

Staff line detection algorithms can be used as a

ﬁrst step in many staff removal algorithms. To un-

derstand the potential of our algorithm to leverage

the performance of existing staff removal algorithms,

we conducted a series of experiments, comparing the

original version of a staff removal algorithm with the

modiﬁed version of it, making use of the Stable Paths

algorithm at the staff line detection step. The quanti-

tative comparison of the different algorithms is totally

in line with the comparison presented in (Dalitz et al.,

2008).

With respect to the considered distortions, regard-

ing the detection stage, the Stable Paths based ap-

proach outperforms the Dalitz algorithm. In Figure 6

we present our results for the removal algorithms:

LineTrack Height (with Dalitz and Stable Paths),

Skeleton and LineTrack Height Modiﬁed (with Stable

Paths). We chose the methods that present the best re-

sults in (Dalitz et al., 2008), implementing our own

removal algorithm with LineTrack Height as a basis.

In general we verify that the replacement of the Dalitz

method by our Stable Paths Approach algorithm as

the staff detection step has improved the ﬁnal staff

line removal results

. Additionally, the LineTracking

Height Modiﬁed algorithm presents an overall better

performance than the original LineTrack Height algo-

rithm from (Dalitz et al., 2008). Our staff line de-

tection and removal approaches also outperform the

Skeleton method, although it continues to present a

competitive performance. We have not integrated the

Stable Paths algorithm with the Skeleton algorithm

as the second performs the lines detection along with

their removal instead of using two separate stages. All

the parameters, both on the Stable Paths detection al-

gorithm and LineTrack Height Modiﬁed, were pre-

liminary tuned over an independent set of images.

This performance gain is even more noteworthy as

the MusicStaves algorithms are receiving as input the

correct number of lines per staff. Had not this been the

case, the differential between both would have been

much larger. In summary, these experiments show the

strength of the algorithms presented here. Despite be-

ing based on simple and intuitive underlying princi-

ples, the performance of the proposed algorithms is

quite competitive.

The analysed results have covered the detection

For the deformations not shown, the stable path is not

signiﬁcantly better than Dalitz.

and removal accuracy but a brief word on speed is also

in order. Comparing different algorithms for speed

is notoriously difﬁcult; we are simultaneously judg-

ing mathematical properties and speciﬁc implementa-

tions. In the experimental study, the current imple-

mentation of the Stable Paths algorithm run almost

as fast as the Dalitz algorithm (20% slower). In re-

spect to the removal algorithms our LineTrack Height

version with Stable Paths is signiﬁcantly faster than

the Skeleton algorithm (two times faster). Compar-

ing to the original LineTrack Height algorithm with

the Dalitz detection algorithm the runtime difference

is not signiﬁcant. The algorithms were evaluated as

available at the Staff Removal Toolkit (Dalitz et al.,

2008).

5 CONCLUSIONS

This paper presented the integration of our robust Sta-

ble Paths Approach algorithm (Cardoso et al., 2008a;

Cardoso et al., 2008b) in the MusicStaves Toolkit

(Dalitz et al., 2008) as a C++ plugin, an improved

version of an existing staff line removal algorithm—

LineTrack Height (Dalitz et al., 2008), and the re-

sults we have obtained in our staff line removal tests.

We have integrated our detection algorithm with ex-

isting staff line removal algorithms. Our approach

successfully deals with the difﬁculties posed by the

symbols superimposed on the staff lines as well as a

wide range of image conditions (e.g., discontinuities,

curved lines), frequently found on handwritten scores.

The encouraging results lead us now to consider

investigating the detection of music symbols bene-

ﬁting from the improved staff line detection and re-

moval, creating a complete OMR application in or-

der to integrate it on our proposed complete OMR

solution—OMRSYS (Capela et al., 2008). Thus, our

proposed system offers a complete solution for the

preservation of our musical heritage. It includes an

optical recognition engine integrated with an archiv-

ing system and a user-friendly interface for search-

ing, browsing and edition. The digitized scores are

stored in MusicXML, a recent and expanding music

interchange format designed for notation, analysis, re-

trieval, and performance applications.

Our proposed algorithms and complete OMR sys-

tem promote the creation of a full corpus of mu-

sic documents, promoting its preservation and study.

This project will culminate in the creation of a repos-

itory of handwritten scores, accessible online. The

database will be available for enjoyment, educational

and musicological purposes, thus preserving this cor-

pus of music in an unprecedented way.

SIGMAP 2008 - International Conference on Signal Processing and Multimedia Applications

268

(a) Curvature. (b) Curvature. (c) White Speckles.

(d) Rotation. (e) Stafﬂine Y-Variation. (f) Stafﬂine Y-Variation.

(g) Stafﬂine Interruptions. (h) Typeset Emulation Part 1. (i) Typeset Emulation Part 2.

Figure 6: Effect of different deformations on the overall staff removal error rates. See (Dalitz et al., 2008) for parameter

details.

ACKNOWLEDGEMENTS

This work was partially funded by Fundac¸

ao para

a Ci

encia e a Tecnologia (FCT) - Portugal through

project PTDC/EIA/71225/2006.

REFERENCES

Capela, A., Cardoso, J. S., Rebelo, A., and Guedes, C.

(2008). Integrated recognition system for music

scores. In International Computer Music Conference

(ICMC) (Accepted).

Cardoso, J. S., Capela, A., Rebelo, A., and Guedes, C.

(2008a). A connected path approach for staff detec-

tion on a music score. In International Conference on

Image Processing (ICIP 2008) (Accepted).

Cardoso, J. S., Capela, A., Rebelo, A., and Guedes, C.

(2008b). Staff detection with stable paths. Trans-

actions on Pattern Analysis and Machine Intelligence

(TPAMI) (Submitted).

Classical Sheet Music Collection (2008). Classical Sheet

Music and MIDI Files, http://www.music-scores.com.

Dalitz, C., Droettboom, M., Pranzas, B., and Fujinaga, I.

(2008). A comparative study of staff removal algo-

rithms. IEEE Transactions on Pattern Analysis and

Machine Intelligence, 30(5):753–766.

Gamera Framework (2008). Gamera Framework,

http://ldp.library.jhu.edu/projects/gamera/.

IOA (2008). Institute of the Arts,

http://patrimonio.dgartes.pt/?lang=pt.

Lester Levi Collection (2008). The Lester

S. Levi Collection of Sheet Music,

http://levysheetmusic.mse.jhu.edu/.

STAFF LINE DETECTION AND REMOVAL WITH STABLE PATHS

269

MacMillan, K., Droettboom, M., and Fujinaga, I. (2002).

Gamera: Optical music recognition in a new shell. In

International Computer Music Conference (ICMC),

pages 482–485.

MIC (2008). Music Information Center, http://www.mic.pt.

MusicStaves Toolkit (2008). MusicStaves Toolkit,

http://lionel.kr.hsnr.de/ dalitz/data/projekte/stafﬂines/.

Mutopia Collection (2008). The Mutopia Project,

http://sca.uwaterloo.ca/Mutopia.

Randriamahefa, R., Cocquerez, J. P., Fluhr, C., Ppin, F., and

Philipp, S. (1993). Printed music recognition. In Pro-

ceedings of the 2nd International Conference on Doc-

ument Analysis and Recognition (ICDAR’93), pages

898–901.

Rebelo, A., Capela, A., da Costa, J. F. P., Guedes, C., Carra-

patoso, E., and Cardoso, J. S. (2007). A shortest path

approach for staff line detection. In AXMEDIS ’07:

Proceedings of the Third International Conference on

Automated Production of Cross Media Content for

Multi-Channel Distribution, pages 79–85, Washing-

ton, DC, USA. IEEE Computer Society.

SIGMAP 2008 - International Conference on Signal Processing and Multimedia Applications

270