Optical character recognition is the procedure by which the computer converts printed materials into ASCII files for editing, compact storage, fast retrieval, and for other purposes. In this study, a neural network approach was applied to perform high accuracy recognition on music score. There are three parts in the study. The first part designed an image preprocessor to obtain sets of images of 17 musical symbols consisting of notes and other glyphs for down-sampling using 8 different digitization procedures to serve as inputs. The second part is composed of the design and implementation of a neural network using a feed-forward with backward propagation. The network was trained using digitized downsampled images of the symbol. The trained network was saved and tested with sample musical symbols for recognition. There were recognition errors. The recognized data was presented to the third part, which was designed as a Cakewalk Application Language (CAL) script generator. CAL scripts were obtained and presented to Cakewalk Audio Pro for Musical Instrument Digital Interfaces (MIDI) production.
[1]
S. E. George,et al.
Scores for Improved Recognition DEFACING MUSIC SCORES FOR IMPROVED RECOGNITION
,
2004
.
[2]
Pierfrancesco Bellini,et al.
Optical music sheet segmentation
,
2001,
Proceedings First International Conference on WEB Delivering of Music. WEDELMUSIC 2001.
[3]
Neill W Campbell,et al.
Sixth International Conference on Image Processing and its Applications
,
1997
.
[4]
Hadar I. Avi-Itzhak,et al.
High Accuracy Optical Character Recognition Using Neural Networks with Centroid Dithering
,
1995,
IEEE Trans. Pattern Anal. Mach. Intell..
[5]
David Bainbridge,et al.
Dealing with superimposed objects in optical music recognition
,
1997
.