27.01.2015 Views

Audiveris: Optical Music Recognition

Audiveris: Optical Music Recognition

Audiveris: Optical Music Recognition

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

MUSIC HACK DAY, Vienna 2013<br />

<strong>Audiveris</strong><br />

<strong>Optical</strong> <strong>Music</strong> <strong>Recognition</strong><br />

Presented by Hervé Bitteur<br />

herve.bitteur@audiveris.org February 2, 2013<br />

V1


Agenda<br />

● User demo<br />

● The big picture<br />

● <strong>Audiveris</strong> technology<br />

● Current 4.2 release<br />

● Integration<br />

● Future<br />

VIENNA 2013<br />

www.audiveris.org<br />

2


User Demo<br />

● User oriented<br />

<br />

<br />

<br />

● Case<br />

Input<br />

Corrections<br />

Outputs<br />

Input : Dichterliebe01.pdf<br />

Language : DEU<br />

Tempo : 80<br />

Output : Finale Notepad<br />

VIENNA 2013<br />

www.audiveris.org<br />

3


Score Transcription « big picture »<br />

<strong>Optical</strong> <strong>Music</strong> <strong>Recognition</strong><br />

Printed<br />

Scores<br />

Scan<br />

Scores<br />

Images<br />

OMR<br />

<strong>Audiveris</strong><br />

Edition<br />

Edit<br />

View<br />

Play<br />

Transpose<br />

MuseScore<br />

Pivot Format<br />

Symbolic<br />

Notation<br />

<strong>Music</strong>XML<br />

<strong>Music</strong><br />

Info<br />

Retrieval<br />

PeachNote<br />

Index<br />

Search<br />

VIENNA 2013<br />

www.audiveris.org<br />

4


<strong>Music</strong> notation<br />

● <strong>Music</strong>XML<br />

Symbol level<br />

High level<br />

Meant for recorded music<br />

Meant for score interchange<br />

Note level<br />

Meant for digital instruments<br />

Sound level<br />

Low level<br />

● MIDI<br />

● MP3<br />

VIENNA 2013<br />

www.audiveris.org<br />

5


OCR vs OMR<br />

Complexity<br />

OCR<br />

(text)<br />

Horizontal lines of<br />

characters<br />

OMR<br />

(music)<br />

Combined horizontal &<br />

vertical directions<br />

Technology Rather mature Several years<br />

behind OCR<br />

Products Many Just a few.<br />

One FOSS: <strong>Audiveris</strong><br />

VIENNA 2013<br />

www.audiveris.org<br />

6


Basic definitions: Runs & Sections<br />

Black run of 3 pixels<br />

Section of 6 adjacent runs<br />

White run of 18 pixels<br />

VIENNA 2013<br />

www.audiveris.org<br />

7


Tech Demo<br />

● Technology oriented<br />

<br />

<br />

Runs<br />

Grid<br />

● Case<br />

<br />

<br />

<br />

<br />

Input : Chula.png<br />

Language : FRA<br />

Params : showRuns, showCombs<br />

Output : MuseScore<br />

VIENNA 2013<br />

www.audiveris.org<br />

8


Pixels: Staff Lines & other objects<br />

VIENNA 2013<br />

www.audiveris.org<br />

9


Black runs histo. → line thickness<br />

VIENNA 2013<br />

www.audiveris.org<br />

10


White runs histogram → interline<br />

VIENNA 2013<br />

www.audiveris.org<br />

11


Vertical runs > line thickness<br />

VIENNA 2013<br />

www.audiveris.org<br />

12


Vertical runs


Short horizontal runs put aside<br />

VIENNA 2013<br />

www.audiveris.org<br />

14


Long hori. runs → lines skeleton<br />

VIENNA 2013<br />

www.audiveris.org<br />

15


Clusters of equidistant lines<br />

VIENNA 2013<br />

www.audiveris.org<br />

16


Staff lines detected<br />

VIENNA 2013<br />

www.audiveris.org<br />

17


Staff lines removed<br />

VIENNA 2013<br />

www.audiveris.org<br />

18


Wavy staff lines: Splines<br />

VIENNA 2013<br />

www.audiveris.org<br />

19


Wavy staff lines: Dewarped image<br />

VIENNA 2013<br />

www.audiveris.org<br />

20


Stems detected<br />

VIENNA 2013<br />

www.audiveris.org<br />

21


Sections → Glyphs → Shapes<br />

VIENNA 2013<br />

www.audiveris.org<br />

22


Generated score<br />

VIENNA 2013<br />

www.audiveris.org<br />

23


<strong>Audiveris</strong> version 4.2<br />

● Released in December 2012<br />

● Features:<br />

<br />

<br />

<br />

<br />

<br />

Ghostscript integration<br />

– for PDF input<br />

Adaptive Binarization<br />

– for non-uniform images<br />

Tesseract OCR V3<br />

– for text handling<br />

ART-based moments<br />

– for shape recognition<br />

Plugins<br />

– for <strong>Music</strong>XML consumers<br />

VIENNA 2013<br />

www.audiveris.org<br />

24


Local OMR Integration<br />

client<br />

scan<br />

display<br />

plugin<br />

<strong>Audiveris</strong><br />

<strong>Music</strong>XML<br />

plugin<br />

MuseScore<br />

VIENNA 2013<br />

www.audiveris.org<br />

25


OMR as a Service (hack of the day)<br />

client<br />

server<br />

display<br />

scan<br />

internet<br />

Front<br />

End<br />

<strong>Music</strong>XML<br />

MuseScore<br />

<strong>Audiveris</strong><br />

VIENNA 2013<br />

www.audiveris.org<br />

26


<strong>Audiveris</strong> evolution<br />

● Technology<br />

<br />

<br />

<br />

Good building blocks available<br />

4.2+: iterative approach OK for good-quality scans<br />

5.x: fuzzy logic needed for poor-quality scans<br />

● Architecture<br />

<br />

Focus on OMR service<br />

● « Freemium » model<br />

<br />

<br />

Free for basic features or low volume<br />

Premium for more serious usage<br />

VIENNA 2013<br />

www.audiveris.org<br />

27


Thank you<br />

www.audiveris.org<br />

<strong>Audiveris</strong> [latin] := « you will have heard »<br />

herve.bitteur@audiveris.org

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!