線上樂譜手寫辨識系統

Abstract

本篇論文中我們提出了一個線上的樂譜手寫辨識系統。樂譜是用來記錄樂曲的工具，作曲家常用其於創作與交流音樂。論文中，我們使用與在紙上相同的書寫方法，利用多筆劃組合出音樂符號。而筆劃的特徵有三種，高度，組成之基本圖形以及方向，可用來辨識出此筆劃所屬類型，再將其組合成所需要的音樂符號。本系統支援基本創作需要之全部音樂符號，辨識率為98.35%，並且提供方便及完善的樂譜修改功能。

In this thesis, we present an online handwritten system for music score recognition. Music score is used to record a music song. People often used to compose a music score on the sheet of paper. In our system, we propose the pen based writing method and use multi-strokes to form a music notation. We extract the height, shape and direction from a stroke as the features and recognize it as a symbol. Then the symbol is combined with other symbols to form a music notation. The system is robust for a general use and supports enough music notations for composition. The recognition rate is 98.35%.