Development of music manuscript database which is searchable by their contents

Research Project

Project/Area Number	17K12795
Research Category	Grant-in-Aid for Young Scientists (B)
Allocation Type	Multi-year Fund
Research Field	Library and information science/Humanistic social informatics
Research Institution	Ritsumeikan University
Principal Investigator	Niitsuma Masahiro 立命館大学, 情報理工学部, 助教 (50733135)
Research Collaborator	Demachi Hazuki Yanagisawa Yoshihiro
Project Period (FY)	2017-04-01 – 2019-03-31
Project Status	Completed (Fiscal Year 2018)
Budget Amount *help	¥2,340,000 (Direct Cost: ¥1,800,000、Indirect Cost: ¥540,000) Fiscal Year 2018: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000) Fiscal Year 2017: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Keywords	OMR / 筆跡鑑定 / 音楽学 / 画像
Outline of Final Research Achievements	Generally speaking a large number of training data is required for DNN applications. It seems the currently available dataset of handwritten musical scores are not sufficient to achieve results satisfying musicologists. Thus, we created more practical database of handwritten musical scores using Yoshitake Kobayashi’s database. Moreover, we proposed a musical symbol isolation method without any preprocessing such as binarization, noise reduction, or staff-line removal, using DNN. In the experiment, more than 90% accuracy rate was achieved for foreground and background separations, but the separation of each musical symbols could not be solved enough. We also experimented with the writer identification using CNN only to achieve 60% for writer identification from dataset containing Johan Sebastian Bach, and his relatives and students, and to prove the difficulty of the task of writer identification from the dataset containing Bach’s copyists such as Christian Gottlob Meissner.
Academic Significance and Societal Importance of the Research Achievements	実際に音楽学者に用いられている古楽譜を考慮したデータベースを築くことで、理想的な画質データを用いた学術的な評価にとどまる研究を抑制する。また、音楽学者が評価を随時行ったことで、実際に音楽学者の利用用途を視野に入れた研究を促す。これまで音楽学者が手探りで行っていた画像処理による画質改善や音符の抽出を解決し、音楽学者が高度な推論に集中するための第一歩を踏み出した。