Evaluation in the Visual Perception of Music

Abstract

This chapter is about the need to evaluate the recognition performed by (i) Optical Music Recognition (OMR) systems, and also by (ii) counterpart on-line systems that directly recognize handwritten music input through a pen-based interface. It presents a summary of reviews that have been performed for commercial OMR systems and addresses some of the issues in evaluation that must be taken into account to enable adequate comparison of recognition performance. A representation language [HEART (HiErARchical Text-Based Representation)] is suggested, such that the semantics of music is captured (including the dynamics of handwritten music) and, hence, a target representation provided for recognition processes. Initial consideration of the range of test data that is needed (MusicBase I and II) is also made. The chapter is motivated by the outstanding need for (i) a greater understanding of how to evaluate the accuracy of music recognition systems, (ii) a widely available database of music test data (potentially automatically generated), (iii) an expression of this test data in a format that permits evaluation (for OMR and on-line systems) and (iv) the proliferation of representation languages — none of which captures the dynamics of handwritten music.