When an OMR system adds a measure where there is none in the ground truth, it creates a problem for automated assessments. The tables for my first blog entry were assembled my hand: I had recognized that OMR2 had split measure 43, and ignored the added measure. This was a human decision, and human decisions do not scale well.
So, can the discovery of added measures be automated? The following table shows a summarized comparison result:
When comparing the number of measures between the ground truth and the OMR results it becomes apparent, that something is off. From the table above we also see that starting with measure 43, every measure has at least two errors. This is a strong indication that measure 43 was split by OMR2. Merging measure 43 44 leads to the following results (summary and details):
So yes, there is a reason to believe that the discovery of added measures can be automated too. Even more, it may even be possible to automate actions like merging measures as well. Of course, the discovery algorithm will need to be much smarter that what is indicated here.