1. Introduction

Computers are used to manipulate music in various forms, for example digital sound recordings, digitized images of printed scores or music representational language (M.R.L.) encodings. This work is concerned with producing M.R.L. data automatically from existing printed music scores.

Computers are commonly used in the manipulation of text and graphics, whether it be creating original work or editing and analysing existing material. The distinction must be made between an image of the material and a representation of the information contained within that image. For instance, a facsimile machine transmits a bit-map image of a page without having any 'knowledge' of the meaning of the symbols or graphics on that page. In contrast, a text document produced by a word-processor is stored as an encoding of the characters used. This enables manipulation of the letters and words involved, for example by globally searching for and replacing one term with another.

Similarly, printed music can be manipulated by the computer in two forms, i.e. a bit-map ('electronic photograph') or an encoding using a music representational language. The former can be used to simply store and retrieve a digitized version of the original printed page, whilst also enabling the use of digital image processing techniques, perhaps to enhance or scale the image. An encoded version of the music printed on the original page can, however, be manipulated in various ways. Thus,


operations such as editing or automatic production of parts from a score are made possible, given the appropriate software. Further possibilities include the production of braille scores or new editions in conventional notation, musicological analysis and electronic publishing.

The above operations are, in fact, equivalent to those now performed on text. The transformation of a small proportion of the enormous amount of existing music into music representational language data by any means other than an automatic system would, however, be an impractical task. The possibility of using a computer to encode the music automatically makes the process possible, although admittedly still a large undertaking.

The text of this thesis is arranged as follows:-

Chapter 2 presents an overview of work undertaken in the field of manipulating printed music by computer. This includes an examination of the various methods available for entering into the computer the information contained in existing scores and a study of the techniques involved in storing the data. A summary of hardware available for the printing of music and a table of significant software packages for printing music are also included. The chapter is a revised and updated version of a previously published paper [Carter 1988]. The discussion of printing hardware has been revised to take account of recent developments in xerographic printers and the increase in significance of the Postscript page description language (P.D.L.). Also, the table of music printing systems has been modified to


include some new entries and increase the information pertaining to systems which have increased in importance.

Chapter 3 provides background information on the subject of pattern recognition, concentrating specifically on the processing of binary images such as engineering drawings, circuit diagrams and flowcharts. A review of past work on pattern recognition of printed music scores is included, which examines the specific problems involved in dealing with this particular class of binary image. The background is thus provided for an in-depth examination of the author's system for automatic production of M.R.L. data from existing printed music scores.

Chapter 4 contains such an analysis. The process from digitized image to M.R.L. data is broken down into its constituent parts, including filtering, staveline-finding, object (symbol) formation and recognition. Illustrations of the various stages are provided.

Chapter 5 contains worked examples of a more substantial nature showing the application of the above system, again, with appropriate illustrations. These examples form the basis of a discussion which covers various aspects of the music recognition problem in detail, including the location of the stavelines when these are substantially obscured, processing handwritten material and dealing with a multi-stave system. Suggestions for solutions to the problems which have arisen are given.

Chapter 6 summarises the qualities of the system described in


chapters 4 and 5 and describes the equipment used for the development work. Some of the possibilities for the future of the system are outlined.