Bibliography on Optical Music Recognition

Optical music recognition is a complex problem because of the stave lines which link up the musical symbols. Thus, the standard approach of segmentation attempts to remove them, without cutting the symbols, by developing a complex algorithm or using them to operate on the zones they delimit. For the segmentation phase, some workers have tried considering a musical score as a set of line segments, but this approach is not appropriate to easy recognition. Our original approach to the segmentation problem is based on an exploitation of these lines and on the creation of virtual interlines according to the musical significance that they have. To limit the influence of the skew, line detection only appears at bar level. The same is true of the final recognition phase which can also deal with a local musical context local for each bar to render the identification. More accurate these lines and virtual interlines define the limits of 2*13 vertical projections (F and G clef) to detect the presence of musical information (patterns). These patterns are coded with 3 characters according to their forms and their relative height and width. A pattern position analysis combined with the score writing rules allow strings of chars matching 2 classes of musical symbols to be created (horizontal analysis to create slurs and vertical analysis for the others). These strings feed a syntactic analyzer for the final recognition according to their musical significance. The checking of the number of beats played in association with the tone of the bar and the identified alterations enables us to guarantee an optimal musical symbols recognition (5 Refs.) segmentation; pattern recognition

The automatic recognition system for printed music would make practical the conversion of large quantities of printed music into computer-readable form. Works on automatic recognition of printed music have been started in the late 1960s and early 1970s. And some recognition systems for music notation have been developed in Japan and Europe. This work presents an overview of works undertaken in the field of manipulating printed music by computer, and proposes a new method to recognize each music symbol using skeletal lines and run-length information. And the system for automatic recognition of printed piano music using the new method has been developed and tested. The result reveals 98.5% accuracy for music symbols. The system shows size-independent, noise immune, and rotation-independent properties (28 Refs.) recognition

Several attempts have been made to implement a computer based automatic music recogniser over the last twenty years. However, for tasks such as music printing or compiling a musical database, no such system has yet become commercially available that would help enter the music into the computer. This paper describes problems that have been ignored by the existing research and which the authors have encountered during the development of a music recognition system. It is argued why solutions to these problems will have to be found before an optical character recognition scheme for music can become reliable enough to be regularly used (7 Refs.) management systems; music; optical character recognition; printing

This thesis deals with Optical Document Recognition. In this domain, reliability is an important point: the user should not have to proofread the whole document. Reliability can be obtained by first improving the quality of the recognition - mostly by solving segmentation problems - and, second, by having the system itself detect badly recognized zones. To fullfill these objectives, one must use a priori knowledge to solve segmentations problems and to represent redundancy, thus allowing error detection. We chose in this thesis to work on Optical Music Recognition for its structured knowledge and because it has many unsolved segmentation problems (mainly linked to information density). For this kind of document strongly syntaxed, we suggest a new method called DMOS (Description and MOdification of Segmentation). It consists of a grammatical formalism of position (to define knowledge) and a parser allowing a dynamic modification of the parsed structure. This modification allows us to introduce context (symbolic level) in segmentation (numeric level) in order to improve recognition. With knowledge represented by a grammar, the DMOS method offers a separation between knowledge and program, and an automatic parser generation (through a compilation phase). These two points greatly ease the management of complex knowledge. Using the formalism of position we have been able to define a grammar of the music notation. The present system is already able to recognize some full scores with polyphonic staves by correcting some type of segmentation errors (like symbols touching notes) and by pointing out badly recognized zones.

In this paper we will show our system for the automated reading of musical scores. The system, as of now, is made up of two modules. The first module takes as input a scanner image. From the image, an intermediate alphanumeric output is obtained. The format we have chosen for this intermediate data is the MUSICA language [De Biasi et al., 1982]. The second module of the system translates the MUSICA code into a standard MIDI file, suitable for immediate playing as well as for further processing (5 Refs.) scanners; optical character recognition

The recognition of sheet music by computer has been tackled with varying degrees of success for 25 years. An effective solution to this problem is to use the model-based approach, where the syntax and structure of sheet music aid the recognition of the music symbols. The complete system, which translates a graphical sheet of music to a text file, involves data acquisition (a flat bed scanner with a resolution of 300 dots per inch is used), bar identification, stave line extraction, symbol segmentation and recognition, and generation of the output file (8 Refs.) model-based reasoning; optical character recognition

Musical data entry methods by typewriter, piano, tablet touch pad, OCR and automatic digital musical analysis are outlined. As examples of musical databases the Institut de Recherche et Coordination Acoustique/musique and the National Ethnographic Museum (of Japan) methods are discussed (9 Refs.) optical character recognition; typewriters

Computer applications for music production have been developed in the 1970s. Computers have been used in various fields of music, including applications for a music database, music CAI and visual input of musical information as well as music production tools. Music as the objective of cognitive science or artificial intelligence has been attracting great interest and international conferences have been held recently. The paper describes the technology and development trends in this field, referring to the basic technology of computer music, notation system, man-machine interaction system, music composition/arrangement systems and music recognition (16 Refs.) instruction; electronic music; music

This paper proposes a new method that improves the note recognition accuracy of a music recognition system. The system recognizes the pitch and the name of the instrument for each note in monaural music signals. The basic idea of the method is utilizing transition pattern of notes in music. Firstly, the results of statistical analyses of printed music are described. These have been conducted in order to obtain the note transition probabilities. Then the method for integrating such probabilistic information into recognition processes is introduced. The method has been implemented using the OPTIMA architecture and tested from several aspects. The results of the tests show that the proposed method improves the note recognition accuracy (6 Refs.)

Page 1. Introducing Feedback into an Optical Music Recognition System

Formal models of music have been used as a design principle of automatic music recognition systems. These models range from statistical to generative/transformational models. However, the use of formal models in recognition problems implicitly impose data structures and algorithmic processes to the mechanism of perception and can result in inconsistent or extremely limited systems. By first modeling the perception mechanism as a statistical decision process merging sensory data and memory information, one can more readily identify the data elements and computational structures required for the recognition task and integrate formal representation(s) of music in a dynamically changing system (26 Refs.) music; statistics

A method for recognizing SYAKUHACHI scores with embedded symbols is proposed. A SYAKUHACHI score mainly consists of characters indicating pitch and embedded symbols indicating the length of the sound. The complementary similarity measure enables the system to handle embedded symbols independently and that results in high performance of the recognition (6 Refs.)

Handwritten character recognition has been increasing in importance and has been expanding its application areas. In this paper, the authors propose a recognition system of a musical score. The recognition system uses a simple, fast feature extraction method, and an efficient way of combining a fast recognition method (3-dimensional table method) and an exact recognition method (multiple simularity method). The authors demonstrated the performance of the completed system by experiments using real data (8 Refs.) transforms; optical character recognition