“A​ ​System​ ​that​ ​produces​ ​MusicXML​ ​File​ ​from​ ​a​ ​Music​ ​Sheet

dealing​ ​with​ ​polyphony​ ​music​ ​arrangements​ ​in​ ​Image​ ​Format”

Abstract
This study presents a system that will improve the generation of MusicXML file from a music
sheet image by introducing a new texture of music, Polyphony, that is two or more simultaneous
lines of independent melody in a musical sheet. This study belongs to deep learning that uses
convolutional​ ​neural​ ​network​ ​for​ ​image​ ​classification.

Introduction

Background​ ​of​ ​the​ ​study

Music has been part of our lives for many years, an art form where efforts of millions of
people have made contributions to it’s evolutions where it created different kinds of forms and
purposes spread throughout the world. It connects the people in different parts of the world with
different cultures and lifestyles. Since music is universal, people all around the world considers
it a symbol of unity where music has the capability to create emotional reactions and these
emotions [1], which are intense feelings that are directed at someone or something [2], are
possessed​ ​by​ ​every​ ​person​ ​in​ ​this​ ​world​ ​making​ ​music​ ​a​ ​very​ ​important​ ​factor​ ​to​ ​the​ ​society.

To understand music is to also understand our society that is why several studies in
music have been conducted in order to deeply understand and discover the components of
music. This is where music theory comes in. Music theory is the study of possibilities and
practices of music. Music theorist attempts to explain the compositional techniques composers
use​ ​in​ ​music​ ​by​ ​establishing​ ​rules​ ​and​ ​patterns[3].

Theoreticians often used music sheet to guide them in conducting their studies since
music sheet is a form of musical notations that embodies the properties of a musical piece like
pitch and duration [4] which for them is a need to know to analyze music properly. In studying
music, a music sheet plays an important role because it serves as a visual representation of a
musical piece. A music sheet can also be used to guide musical performances and keep them
from properly executing the music that they are playing. Musicians also used music sheet to
create and compose new music. A music sheet can be also used by common people who just
wants to learn music as long as they know how to to read and understand it. Music sheet is very
important in music because of all great uses it can offer but without the ability of reading the
music sheet it would be useless for us that is why being able to read music sheet is very
important.
 In music theory, one of the basic elements of music is texture [5]. Texture describes how
much is going on in the music at any given moment, it might be made up of rhythm only, or of a
melody line with chordal accompaniment, or many interweaving melodies [5]. To formally
describe texture in music, these four terms, namely Monophony, Homophony, Heterophony and
Polyphony​ ​are​ ​used​ ​[5].​ ​The​ ​study​ ​that​ ​will​ ​be​ ​conducted​ ​will​ ​focus​ ​on​ ​Polyphony​ ​specifically.

Since the study will focus on polyphonic music there is a need to understand how
polyphony is applied in music sheet and this is where musical notation comes in. Identifying the
difference between the musical notations present in a monophonic music sheet versus
polyphonic music sheet should be done. Since polyphony is more complex in nature because of
the multiple melodies it comprise [5] against the monophy which comprises only one melody it
would​ ​be​ ​given​ ​that​ ​the​ ​polyphonic​ ​music​ ​sheet​ ​have​ ​more​ ​properties​ ​of​ ​musical​ ​notations.

There have already been existing studies that were conducted that focused on
monophonic music that is why the study that will be conducted chooses Polyphonic texture. The
system will aim to improve the existing study which is to generate a MusicXML file from a music
sheet​ ​image.

Statement​ ​of​ ​the​ ​Problem

Every person in this world can appreciate music but learning music would require a
special dedication. In learning music, reading a music sheet is a requirement and in order to do
that it needs to take time in learning the musical notations which can be very tedious. Although
some have the abilities to learn faster, most of the people can have a very hard time learning it.
All music can be represented by a music sheet, therefore people who can understand it have an
advantage​ ​of​ ​learning​ ​music​ ​better​ ​which​ ​also​ ​means​ ​a​ ​better​ ​appreciation​ ​of​ ​music.

With the advancement of technology applied in music, there have been an incline growth
of softwares that enables music sheet to be read in computers which now becomes a “digital
music sheet” and can be converted in different virtual formats. In most cases, it can be a MIDI
file, an audio (wav, mp3), or a MusicXML export. In those formats, MusicXML, a digital music
sheet interchange and distribution format, has many programs/softwares in the market that
make​ ​use​ ​of​ ​it.

Interest in Music Information Retrieval (MIR) is recently very increasing and large
amount of musical data is available in the internet [6] because of digitalization of these musical
works. Since digitized images will be used in the system it would then allow to extract
information further by using the music sheet images and extract the musical notations through
the​ ​generation​ ​of​ ​MusicXML​ ​format​ ​thereby​ ​preserving​ ​these​ ​valuable​ ​informations.
 Virtually almost all of the music we listen today is polyphonic in texture [7] which means
music have more than one melodies playing at a time. Given that polyphony is intrinsic with
music nowadays, it’s equally important that we need to understand the musical notation
associated with it and apply it through MusicXML. Thus to address the problem, this study will
provide​ ​a​ ​system​ ​that​ ​can​ ​generate​ ​MusicXML​ ​file​ ​from​ ​a​ ​Polyphonic​ ​music​ ​sheet​ ​image.

Objectives​ ​of​ ​the​ ​Study

The objective of the study is to make a system that accepts a music sheet with
polyphonic texture in an image format and generate its equivalent MusicXML file. Specifically,
this​ ​study​ ​aims​ ​to​ ​produce​ ​a​ ​system​ ​that​ ​can:

1. Accept​ ​a​ ​polyphonic​ ​music​ ​sheet​ ​image​ ​as​ ​the​ ​input.

2. Build a deep learning image classifier using convolutional neural network with inception
model​ ​using​ ​transfer​ ​learning.
3. Generate​ ​a​ ​MusicXML​ ​file​ ​equivalent​ ​to​ ​the​ ​music​ ​sheet​ ​image​ ​input.

RRL​ ​-​ ​Review​ ​of​ ​Related​ ​Literature

Introduction

Many musical works produced in the past are still available only as original manuscript
or as photocopies, which includes music sheets. The preservation of these works requires
digitalization and transformation into a machine-readable format. Digitalization has been
commonly used as a possible tool, offering easy duplications, distribution and digital processing
of these information [8]. There are many ways to represent these digital format an example is in
a form of Portable Network Graphics (PNG). Although having these music sheets in digital
format preserves the information, it doesn’t not extract the musical semantics behind those
musical notation of those images which means the only usage of it is just to be read or printed
but​ ​it​ ​can’t​ ​be​ ​played.

Music​ ​Notation​ ​Interchange​ ​Format​ ​(MusicXML)

Before, sharing music between different kinds of music software programs used to be
difficult since no one music program can do everything equally [9]. As the time goes by, music
application in technology was able to develop a music notation interchange format called MIDI.
But even with the invention of MIDI, it was not able to represent music notation since it is only
good​ ​for​ ​performance​ ​application​ ​such​ ​as​ ​sequencers.
 Until people in the community in line with music and technology decided to come up with
new music interchange format such as NIFF and SMDL but these formats have limitations. NIFF
does not work very well for general interchange. It also represent music in a graphical format
which means it doesn’t have a concept of “C” and another major impediment is that it’s in a
binary format which makes it harder to debug [9]. SMDL, on the other hand, became too
complex​ ​to​ ​work​ ​with.

Finally, MusicXML was created by Recordare LLC. MusicXML solved the limitations
other music notation interchange formats have, it was able to characterize the western music
notation. To define clearly, a MusicXML is an interchange format for music notation, music
analysis, music information retrieval, and musical performances [10] which now becomes the
standard for interactive music sheet. MusicXML is used by many applications to perform music
related​ ​task​ ​like​ ​playing​ ​the​ ​music​ ​semantics​ ​inside​ ​the​ ​format.

Software programs have been developed to generate MusicXML such as abc2xml,

Audio Score Ultimate, MuseScore 2, Finale Notepad, and Sibelius. Each of the programs can
generate MusicXML in different ways. The ‘abc2xml’ program can only produce MusicXML by
using a command-line tool and a text input while Audio Score Ultimate transcribes CD and MP3
tracks into musical notation which can be exported to MusicXML. MuseScore 2, originally from
MuseScore, even with its improvements, just like the remaining programs, requires tedious task
in generating MusicXML because these programs will just provide an empty sheet and the user
will have to input the musical symbols he/she wants by dragging it. The two programs ‘abc2xml’
and ‘Audio Score Ultimate’ have different input formats while the remaining three programs
‘MuseScore 2’, ‘Finale Notepad’ and ‘Sibelius’ requires maximum time and energy which is also
error​ ​prone​ ​since​ ​it’s​ ​manually​ ​done​ ​by​ ​the​ ​user.

Therefore, a good way to generate a MusicXML is to provide an input of music sheet

image​ ​to​ ​the​ ​program.​ ​This​ ​way,​ ​time,​ ​effort​ ​and​ ​errors​ ​are​ ​minimized.

MusicXML​ ​Generation​ ​Softwares

Programs have been developed through the years were an image music sheet can be
converted to MusicXML. The most notable programs are Audiveris, Sharpeye and Pdf2Music
and​ ​Musa.

Audiveris
Audiveris is an open-source Optical Music Recognition software. It allows user to import
scanned music sheet and export them into MusicXML format. Although it does what the study
wanted, it have some difficulties in reading music symbols like beam [11]. Programs have been
created to accept MusicXML format and then play it accordingly. A good example of that
program is the system created by Mr. Michael Alan V. Ygnacio in his special problem: “A system
that converts a music sheet with round notes and with shape notes”. The system plays accepts
the MusicXML generated by Audiveris and then with the translated round notes convert to
shape notes. Since the system heavily relies on Audiveris, the reading and rendering of music
sheet image to MusicXML, he can’t modify the ability of Audiveris to accurately scan an image
of a music sheet. Mr. Ygnacio said that “Audiveris has difficulties on reading the music symbols
and​ ​other​ ​data​ ​within​ ​the​ ​scanned​ ​music​ ​sheet”​ ​[11].

Sharpeye
SharpEye is an optical music recognition program which has an interactive graphical
environment for editing the symbolic music notation extracted from scanned music. It can scan
and convert printed music sheet into a musicxml file or a midi file [12]. It only uses bitmap and
tiff images. However, Sharpeye has its drawbacks. According to Vrist, there are missing flags on
the​ ​eighth​ ​notes,​ ​missing​ ​notes​ ​and​ ​missing​ ​flats​ ​or​ ​sharps​ ​[13].

Pdf2Music
Pdf2Music offers pro and non-pro services but both of them doesn’t offer conversion of
images​ ​to​ ​MusicXML. The major drawback in Pdf2Music is that it generates a bad MusicXML
format from the music sheet in pdf format [13]. Missing information and wrong placement of
music​ ​symbols​ ​can​ ​be​ ​seen​ ​in​ ​the​ ​MusicXML​ ​file.

Template​ ​Matching
Template Matching is a high-level machine vision technique that identifies the parts on
an image that match a predefined template. It has two main disadvantages. First, it needs the
template to be stored in a memory for correlation and second, it is sensitive to noise and scale
which​ ​makes​ ​it​ ​prone​ ​to​ ​errors.

Musa
A new system developed by Mr. Ramie Jeanne Cortes in her special problem: “A
System that Produces a MusicXML File From a Music Sheet in Image Format” to solve the
limitations of the previously talked tools. The system generates MusicXML file from a music
sheet image just what the system wanted but one particular limitation she stated in her SP is
that her system only deals with Monophonic single-part arrangements, that is only single note
will be played at a time or solo melody [4]. This limits the possibilities of extracting more
information​ ​form​ ​a​ ​music​ ​sheet​ ​image.
The​ ​Challenge​ ​of​ ​Optical​ ​Music​ ​Recognition
Optical​ ​Music​ ​Recognition​ ​(OMR)​ ​is​ ​the​ ​problem​ ​of​ ​converting​ ​a​ ​scanned​ ​image​ ​of​ ​sheet
music​ ​into​ ​a​ ​symbolic​ ​representation​ ​like​ ​MusicXML​ ​[15]​ ​or​ ​MIDI.​ ​Almost​ ​all​ ​of​ ​the​ ​presented
previous​ ​works​ ​applies​ ​OMR​ ​in​ ​their​ ​system.​ ​In​ ​OMR,​ ​the​ ​most​ ​common​ ​algorithms​ ​it​ ​follows
are​ ​decomposed​ ​into​ ​four​ ​task:
1. Image​ ​Preprocessing
2. Segmentation
3. Object​ ​Recognition
4. Semantic​ ​Reconstruction

Despite the fact that OMR systems have been researched thoroughly over the last few
decades and even several commercial tools exist, the practical results are still far from ideal
[16]. Since image preprocessing is the initial step of OMR, it can affect the next processes and if
something wrong happens in this stage, the recognition would be useless. One of the most
important open issues of the OMR research is the lack of an available ground-truth database
where it could serve as a benchmark [17] and the absence of common methodologies and
metrics that would compare the results of OMR systems [17]. Given all those limitations
regarding​ ​OMR,​ ​the​ ​study​ ​will​ ​find​ ​another​ ​way​ ​that​ ​doesn’t​ ​use​ ​OMR.

Given with all of the presented previous works related to the generation of MusicXML
format from a music sheet image, it is clear that the limitations and shortcomings of these
systems are apparent. This implies that a need for a new system that would generate a
MusicXML file from a Polyphonic music sheet image is needed. By introducing a new system
involving deep learning to identify music semantics in music sheet image would hopefully solve
the​ ​presented​ ​tools’​ ​limitations​ ​and​ ​impediments.

Conceptual​ ​Framework

The​ ​following​ ​are​ ​the​ ​terms​ ​and​ ​concepts​ ​that​ ​are​ ​used​ ​throughout​ ​the​ ​study.

Deep​ ​Learning

Deep learning, also known as Deep Neural Learning or Deep Neural Network, is a
subset of machine learning in Artificial Intelligence (AI) that has networks which are capable of
learning unsupervised from data that is unstructured or unlabeled [18]. Essentially it involves a
lot of input data into a computer system which in return can make decisions about other data.
Using deep learning, music sheet images will be used as the data to be fed in the computer
system. Later on, if the system will be given an input of another music sheet image, it would be
able​ ​to​ ​identify​ ​the​ ​music​ ​semantics.
Convolution​ ​Neural​ ​Network

Convolutional Neural Networks are very similar to ordinary Neural Networks which are
made up of neurons that have learnable weights and biases and each neuron receives some
inputs, performs a dot product and optionally follows it with a non-linearity [19]. Convolutional
neural network are essentially black box that constructs features that can be handcrafted and
these abstract features created from training are so generalized that they account for variance.
Creating a system that has manually trained CNN would take a lot of computing power and a lot
of​ ​time​ ​both​ ​of​ ​which​ ​is​ ​beyond​ ​of​ ​my​ ​capabilities.

Inception​ ​Model

An inception model is a pre-trained CNN Model. Inception model was trained by Google
on 100K images with a thousand categories. The use case of the study will focus on the
classification of music symbols of the polyphonic music sheet image. But inception was not
trained​ ​on​ ​these​ ​music​ ​symbols​ ​that​ ​is​ ​why​ ​we​ ​use​ ​transfer​ ​learning.

Figure​ ​[19]

Transfer​ ​Learning

A process called transfer learning means applying the learnings from a previous training
session to a new training session. Looking at the inception model, above, we can see that when
we feed an image as an input at each layer it will perform a series of operations on that data
until it outputs a label and a classification percentage each layer is a different set of abstractions
in the first layers it's basically taught itself edge detection then shape detection in the middle
layers​ ​and​ ​they​ ​get​ ​increasingly​ ​more​ ​abstract​ ​up​ ​until​ ​the​ ​end.

Looking at the last few layers these are the highest level detectors for whole objects. For
transfer learning, the system will basically just want to retrain that last layer on features of the
music​ ​symbols​ ​so​ ​it​ ​can​ ​add​ ​a​ ​representation​ ​to​ ​its​ ​repository​ ​of​ ​knowledge.
Music​ ​Notations

Stave
The staff (plural ​staves​) consists of five horizontal parallel lines and four spaces. Each of
those lines and each of those spaces represents a different letter, named A-G, which in turn
represents a note and the note sequence moves alphabetically up the staff [20]. In polyphonic
music, two independent melodies are played at the same time which means more than one
notes​ ​are​ ​active​ ​or​ ​played​ ​at​ ​the​ ​same​ ​time.

In the figure below, the music sheet can be determined as polyphonic in texture since
the number of staves are more than one and often times it is enclosed in a ​brace​. Although
there​ ​are​ ​other​ ​cases​ ​where​ ​only​ ​one​ ​set​ ​of​ ​staff​ ​can​ ​be​ ​seen​ ​other​ ​measures​ ​can​ ​be​ ​used.

Figure​ ​[21]

Key​ ​Signatures​ ​&​ ​Clef

The pitch of a note is how high or low it sounds [21]. Key signatures determine pitch by
using sharps and flats. In the study, key signatures will be vital in determining the notes
equivalent value. The clef on the music sheet can be a treble clef or a bass clef. The treble clef
has the ornamental letter G on the far left side and it encircles the “G” line on the staff [20] while
the bass clef is also referred as the F clef [20]. Most music sheet have these important music
notations because clef tells the letter name of the note and the Key tells whether the note is
sharp, flat or natural. Using these notations, we can now identify the notes and label them
properly.

Position​ ​of​ ​Notes​ ​and​ ​Rest

The note’s letter names are identified by the position in the staff. With this, it will be easy
to​ ​locate​ ​the​ ​note​ ​in​ ​the​ ​staff​ ​[22].
 Methodology

This study will implement Image Classification using deep learning through
Convolutional Neural Network using Inceptional Model with the help of transfer learning. These
are the tasks: Data Acquisition, Data Labeling, Image Preprocessing, and generation of
MusicXML. The development process will follow the process of Analysis, Design,
Implementation​ ​and​ ​Testing​ ​phase.

Analysis
The​ ​following​ ​will​ ​be​ ​the​ ​analysis​ ​of​ ​each​ ​OMR​ ​tasks:

- Analysis​ ​on​ ​Data​ ​Acquisition

A collection of music sheet images have to be gathered, a large set is preferable.

Individual properties of the musical sheet, such as the individual notes, should be
extracted.​ ​Using​ ​tools​ ​for​ ​cropping​ ​such​ ​as​ ​snipping​ ​tool​ ​in​ ​windows,​ ​we​ ​can​ ​get​ ​it.

During the process of cropping, all five line of the staff as well as the entire beam
of the note should be visible in the resulting picture. Since the study deals with
polyphonic music sheets, identifying the properties that are unique in polyphonic music is
important. The number of stave and notes played at the same time are identified. An
image ratio should be defined for the extracted images. Below is an example image of a
music sheet. ​(Further studies will be conducted in order to properly classify the music
symbol​ ​images​ ​will​ ​be​ ​conducted)

Figure​ ​generated​ ​using​ ​MuseScore​ ​2

- Analysis​ ​on​ ​Data​ ​Labeling

The consisting data set would then be composed of pictures of the properties
extracted in the music sheet image. Labelling each of those data would then take place.
Identifying each properties according to the music sheet image. Since the study deals
with polyphonic music, labelling the images according to the properties of a polyphonic
music is done. Labelling the notes can be done through identifying the key signature of
the​ ​score​ ​and​ ​the​ ​clef​ ​it​ ​is​ ​using​ ​and​ ​the​ ​position​ ​of​ ​the​ ​note​ ​in​ ​the​ ​staff.

- Analysis​ ​on​ ​image​ ​preprocessing

The digital image which is extracted during the Data Acquisition have different
sizes and ratios. That is why images would have to be the preprocess in order to have
similar​ ​sizes​ ​and​ ​ratios.

- Analysis​ ​on​ ​Classifier​ ​Retraining

Retraining of the classifier with the newly linked Music Symbol Image dataset will
be done to follow the concept of transfer learning. The classifier would then return an
ordered​ ​list​ ​based​ ​on​ ​the​ ​identified​ ​music​ ​symbols​ ​in​ ​the​ ​music​ ​sheet​ ​image.

- Analysis​ ​on​ ​generation​ ​of​ ​MusicXML

The ordered list of music symbols will be converted into MusicXML format. For
generating an output, it must follow the MusicXML Data Type Document Specification.
The​ ​result​ ​should​ ​use​ ​the​ ​tags​ ​uniform​ ​to​ ​all​ ​MusicXML​ ​files.
 REFERENCES

[1]​ ​Vuust,​ ​P.,​ ​&​ ​Kringelbach,​ ​M.​ ​L.​ ​“The​ ​Pleasure​ ​of​ ​Making​ ​Sense​ ​of​ ​Music”.​ ​Interdisciplinary
Science​ ​Reviews​ ​(2010):​ ​166-182.​ ​Accessed​ ​2017

[2]​ ​Lydeen,​ ​L.​ ​F.​ ​(1987).​ ​Emotions​ ​and​ ​moods:​ ​medical​ ​&​ ​psychological​ ​subject​ ​analysis​ ​with
bibliography.​ ​Washington,​ ​D.C.:​ ​Abbe​ ​Association.​ ​Accessed​ ​2017

[3]​ ​M.​ ​(n.d.).​ ​Music​ ​theory​ ​and​ ​classic​ ​harmony.​ ​Retrieved​ ​October​ ​24,​ ​2017,​ ​from
http://www.aboutmusictheory.com/classic-music-theory.html

[4]​ ​Cortes,​ ​R.​ ​and​ ​Roxas,​ ​R.​ ​A​ ​System​ ​that​ ​Produces​ ​a​ ​MusicXML​ ​File​ ​From​ ​a​ ​Music​ ​Sheet​ ​in
Image​ ​Format

[5]​ ​M.​ ​(n.d.).​ ​Music​ ​texture​ ​theory​ ​–​ ​Monophony​ ​or​ ​Polyphony.​ ​Retrieved​ ​September​ ​14,​ ​2017,
from​ ​http://www.aboutmusictheory.com/music-texture.html

[6]​ ​Hutchison,​ ​D.,​ ​Jensen,​ ​K.,​ ​Kanade,​ ​T.,​ ​Kittler,​ ​J.,​ ​Kleinberg,​ ​J.​ ​M.,​ ​Kronland-Martinet,​ ​R.,​ ​.​ ​.​ ​.
Ystad,​ ​S.​ ​(n.d.).​ ​Computer​ ​Music​ ​Modeling​ ​and​ ​Retrieval.​ ​Genesis​ ​of​ ​Meaning​ ​in​ ​Sound​ ​and
Music.​ ​Berlin,​ ​Heidelberg:​ ​Springer​ ​Berlin​ ​Heidelberg.

[7]​ ​MENU.​ ​(2017,​ ​March​ ​30).​ ​Retrieved​ ​November​ ​23,​ ​2017,​ ​from
https://livingpianos.com/music-theory/what-is-polyphony-in-music/

[8]​ ​Paredes,​ ​R.,​ ​Cardoso,​ ​J.​ ​S.,​ ​&​ ​Pardo,​ ​X.​ ​M.​ ​(2015).​ ​Pattern​ ​Recognition​ ​and​ ​Image​ ​Analysis
7th​ ​Iberian​ ​Conference,​ ​IbPRIA​ ​2015,​ ​Santiago​ ​de​ ​Compostela,​ ​Spain,​ ​June​ ​17-19,​ ​2015,
Proceedings.​ ​Cham:​ ​Springer​ ​International​ ​Publishing.

[9]​ ​MusicXML​ ​FAQ.​ ​(n.d.).​ ​Retrieved​ ​November​ ​21,​ ​2017,​ ​from

http://www.musicxml.com/tutorial/faq/

[10]​ ​Introduction.​ ​(n.d.).​ ​Retrieved​ ​November​ ​25,​ ​2017,​ ​from

http://www.musicxml.com/publications/makemusic-recordare/notation-and-analysis/introduction/

[11]​ ​Ygnacio,​ ​M.​ ​and​ ​Roxas,​ ​R.​ ​A​ ​System​ ​that​ ​Converts​ ​Music​ ​Score​ ​Sheets​ ​with​ ​Round​ ​Notes
into​ ​One​ ​with​ ​Shape​ ​Notes.​ ​(pages​ ​4)

[12]​ ​Visiv.​ ​(n.d.).​ ​Retrieved​ ​December​ ​1,​ ​2017,​ ​from​ ​http://www.visiv.co.uk/

[13]​ ​Vrist,​ ​Soren​ ​Bjerregaard​ ​(2009).​​ ​Optical​ ​Music​ ​Recognition​ ​for​ ​Structural​ ​Information​ ​from
high-quality​ ​scanned​ ​music,​ ​p.39,​ ​45
[14]​ ​(n.d.).​ ​Retrieved​ ​December​ ​5,​ ​2017,​ ​from
https://docs.adaptive-vision.com/current/studio/machine_vision_guide/TemplateMatching.html

[15]​ ​ ​J.​ ​Ganseman,​ ​P.​ ​Scheunders,​ ​and​ ​W.​ ​D’haes.​ ​Using​ ​xquery​ ​on​ ​musicxml​ ​databases​ ​for
musicological​ ​analysis.​ ​In​ ​Proceedings​ ​of​ ​the​ ​9th​ ​International​ ​Conference​ ​on​ ​Music​ ​Information
Retrieval,​ ​pages​ ​433–438,​ ​Philadelphia,​ ​USA,​ ​September​ ​14-18​ ​2008.​ ​http://ismir2008.ismir.
net/papers/ISMIR2008_217.pdf.

[16]​ ​Rebelo,​ ​A.,​ ​Fujinaga,​ ​I.,​ ​Paszkiewicz,​ ​F.,​ ​Marcal,​ ​A.,​ ​Guedes,​ ​C.,​ ​Cardoso,​ ​J.:Optical​ ​music
recognition:​ ​state-of-the-art​ ​and​ ​open​ ​issues.​ ​International​ ​Journal​ ​of​ ​Multimedia​ ​Information
Retrieval​ ​1(3),​ ​173–190​ ​(2012)

[17]​ ​Novotný,​ ​J.,​ ​&​ ​Pokorný,​ ​J.​ ​(2015).​ ​Introduction​ ​to​ ​Optical​ ​Music​ ​Recognition:​ ​Overview​ ​and
Practical​ ​Challenges.​ ​DATESO.

[18]​ ​Momoh,​ ​O.​ ​(2016,​ ​October​ ​30).​ ​Deep​ ​Learning.​ ​Retrieved​ ​December​ ​6,​ ​2017,​ ​from
https://www.investopedia.com/terms/d/deep-learning.asp

[19]​ ​Convolutional​ ​Neural​ ​Networks​ ​(CNNs​ ​/​ ​ConvNets).​ ​(n.d.).​ ​Retrieved​ ​December​ ​9,​ ​2017,
from​ ​http://cs231n.github.io/convolutional-networks/#layers

[20]​ ​Says,​ ​R.​ ​L.​ ​(2016,​ ​June​ ​10).​ ​How​ ​to​ ​Read​ ​Sheet​ ​Music:​ ​Step-by-Step​ ​Instructions.
Retrieved​ ​December​ ​13,​ ​2017,​ ​from
https://www.musicnotes.com/blog/2014/04/11/how-to-read-sheet-music/

[21]​ ​Schmidt-Jones,​ ​C.,​ ​&​ ​Jones,​ ​R.​ ​(2007).​ ​Understanding​ ​basic​ ​music​ ​theory.​ ​Houston,​ ​TX:
Connexions.

[22]​ ​Music​ ​Note​ ​Names.​ ​(n.d.).​ ​Retrieved​ ​December​ ​13,​ ​2017,​ ​from
http://www.piano-keyboard-guide.com/music-note-names.html
 UNIVERSITY​ ​OF​ ​THE​ ​PHILIPPINES

Bachelor​ ​of​ ​Science​ ​in​ ​Computer​ ​Science

Charlito​ ​G.​ ​Piao​ ​Jr.

A​ ​System​ ​that​ ​produces​ ​MusicXML​ ​File​ ​from​ ​a​ ​Music​ ​Sheet

dealing​ ​with​ ​polyphony​ ​music​ ​arrangements​ ​in​ ​Image​ ​Format

SP​ ​Adviser:

Robert​ ​R.​ ​Roxas,​ ​PhD

Department​ ​of​ ​Computer​ ​Science

University​ ​of​ ​the​ ​Philippines​ ​Cebu

Date​ ​of​ ​Submission

December​ ​2017