臺灣博碩士論文加值系統

本篇論文針對光學樂譜辨識系統(OMR)提出一套可靠的辨識樂譜系統架構。它可以自動辨識光學掃描後的樂譜影像，並轉換MIDI音樂檔。利用光學掃描辨識樂譜好處如下，建立音樂資料庫減少記憶體空間，易編輯修改或列印，以及產生音樂使表演者可伴奏練習。
我們所採用的技巧如下：霍夫轉換、影像旋轉、垂直與水平正交投影、樂譜相關參數與樂理規則的限制，最後根據MIDI文件格式的自編程式轉換出MIDI音樂檔，以上這些技術會在本論文中詳細說明。
本系統流程分四大階段，第一階段影像前置處理，首先須對影像大小和角度修正至五線譜線呈水平線段，讓後續步驟影像辨識較為簡易；第二階段，使用五線譜與小節線的偵測與移除，使所有音樂符號孤立；第三階段，進行音符辨識，並檢驗小節線節拍是否準確，整合其整體辨識結果，並產生結果數據；第四階段，將辨識結果帶入程式轉換MIDI音樂格式檔，再藉由撥放軟體將旋律播放出來。實驗結果在本篇可迅速辨識各種樂譜圖像，同時效能和其他限制也會在此論文中提出。

In this thesis, we aim at the optical music recognition system (OMR) to propose a reliable identification system. It can automatically identify the music score image after optical scanning processing, and convert MIDI music files. Recognition music image using optical scanning has many benefits to: establish the music library to reduce memory space, to edit easily if need to modify or print, and to produce music sound so that users can accompany exercise.
The techniques we use are as follows: Hough transform, image rotation, vertical and horizontal orthogonal projection, music-related parameters and music theory limit, and according to MIDI file format conversion programs to convert MIDI music files through the self-written programs. These techniques will be described in detail in this thesis.
The system processes has four major phases: at the first stage, the image pre-processing: in order to let process image recognition easily, we must normalize the image so that staff lines will parallel the horizontal axis. The second stage, in order to isolate all music symbols, we use the staff line detection and bar line detection. The third stage, we identify the music notes, test the accuracy of rhythm section line, and final produce the resulting data. The fourth stage, the program will convert the result into the MIDI music file, and then play music through the player software. The experimental results can quickly identify everything variety of music image, while restrictions and effectiveness will be presented in this thesis.

致謝辭 i
中文摘要 ii
英文摘要 iii
目錄 iv
圖目錄 vi
表目錄 viii
第一章 簡介 1
1.1動機 1
1.2音樂樂理知識 1
1.3相關研究 7
1.4論文架構 12
第二章 理論基礎 13
2.1正交投影法 13
2.2霍夫轉換 14
2.3樂理規則 15
2.4 MIDI檔案格式 16
第三章 研究流程 19
3.1影像前置處理 19
3.1.1角度偏斜修正 19
3.1.2 進階影像二元化 21
3.2五線譜與小節線辨識與濾除 22
3.2.1五線譜辨識與濾除 23
3.2.2小節線辨識與濾除 24
3.3音符辨識 26
3.3.1辨識有符幹的音符 26
3.3.2辨識沒有符幹的音符 31
3.3.3小節線檢測 32
3.4格式轉換 35
第四章 實驗結果與分析 37
4.1實驗環境 37
4.2實驗結果 38
4.3優劣性分析 44
第五章 結論與未來展望 47
參考文獻 48
附錄一 50
作者簡介 56

[1] D. Blostein, and H. S. Baird, “A Critical Survey of Music Image Analysis”, Structured Document Image Analysis, Springer Verlag, pp.405-434, 1992.
[2] K. T. Reed, and J. R. Parker, “Automatic Computer Recognition of Printed Music”, Proceedings of the ICPR, pp. 803-7, 1996.
[3] Y. C. Chung, “Recognition of Printed Sheet Music Using Hough Transform and Morpholgy Operations”, Master Thesis, National Taiwan Normal University, Taiwan, 1995.
[4] E. Sicard, “An Effcient Method for The Recognition of Printed Music”, Proceedings of the 11th IAPR, pp. 573-6,1992.
[5] F. Kimura, and M. Shridha, “Handwritten Numerical Recognition Based on Multiple Algorithms”, Pattern Recognition, 19(1), pp 1-12, 1986.
[6] R. J. Randriamahefa, J. P. Cocqurerz, C. Fluhr, and F. Pepin, and S. Philipp, “Printed Music Recognition”, Proceedings of ICDAR, pp.898-901, 1993.
[7] K. Wijaya, and D. Bainbridge, “Staff Line Restoration”, Image Processing and Its Applications, Seventh International Conference, pp760-764, 1999.
[8] I. Leplumey, J. Camillerapp, and G. Lorette, “A Robust Detector for Music Staves”, Document Analysis and Recognition, Proceedings of the Second International Conference, pp. 902-905, 1993.
[9] D. Pruslin, “Automatic Recognition of Sheet Music”, Sc. D. Dissertation, Massachusetts Institute of Technology, 1996.
[10] 林文靖，"樂譜自動辨識之電腦視覺系統設計"，國立中正大學電機工程研究所碩士論文，民國90年7月。
[11] 蔡自偉，"印刷樂譜辨識系統"，國立中山大學資工學系碩士論文，民國93年7月。
[12] 鐘國亮，「影像處理與電腦視覺」，台灣東華書局，書號1201222，民國93年2月。
[13] (.mid) Standard MIDI File Format, http://faydoc.tripod.com/formats/mid.htm.
[14] MIDI 文件格式，http://myfaq.com.cn/A/2001-12-20/8449.html.