Assessing Optical Music Recognition Tools

The Optical Music Recognition task is more complex than OCR. Despite to the availability of several commercial OMRs: SharpEye2, SmartScore, Photoscore, CapellaScan, etc., none of these is satisfactory in terms of precision and reliability. The efficiency declared by the each distributor is close to 90%, but this value is obtained only when quite regular music sheets are processed and the estimation is not always objective. In the character or face recognition field, there are many ground truth databases that enable recognition results to be evaluated automatically and objectively. At the present time, there is neither a standard database for music score recognition or a standard terminology. If a new recognition algorithm or system were proposed, it could not be compared with the other algorithms or systems since the results would have to be traditionally evaluated with different scores and different methods. Taking these facts into consideration, it is indispensable to make a master music score database that can be used to objectively and automatically evaluate the music score recognition system. At the same time a set of rules and metrics are needed in order to define what aspects have to be considered in the evaluation.