A Complete Optical Music Recognition System : Looking to the future

Data Type support. Functional qualities, such as `map' and `apply'. Is Such a System Possible? Work carried out has already proven that many of the key ideas to the proposed system are realistic [Bai91], [Bai94] and [Bai95]. Compelling arguments can be given, using only existing computer science knowledge and techniques, that the remaining concepts are attenable. Dynamic primitives are fully realised in the current system. Geometric shapes speci ed in the drawing package can be directly imported into the primitive identi cation program, and the tolerances for matching can be speci ed dynamically. Currently a text editor is used to alter tolerances. Ideally this functionality should be incorporated into the drawing package, however this is a purely aesthetic issue. In the implemented system, the legal con gurations of musical primitives are de ned dynamically using a specially designed language. Though signi cant musical knowledge can be expressed using the language, it is too simple, verbose at times, and ultimately limited. Qualities for an improved language have been presented earlier in this report. These anticipated requirements do not exceed the mechanisms o er by existing compiler techniques. The existing matching algorithm is naive | a consequence of the simple language | and could not easily be extended to match the powerful descriptions of legal con gurations expressed by the new language. Semantic Networks and Frame-Based Systems would permit sophisticated matching techniques that would equal the power of the proposed language. There is a substantial body of literature involving these two AI techniques. Work in computer vision is particularly relevant. The idea of building an internal data structure representing the musical meaning of the graphically recognised shapes is implemented in the current system. Using this approach it was a simple task to