I use an OCR software to recognize music https://www.neuratron.com/photoscore.htm
it fails miserably if the music sheet is a little off due to bad scanning - works perfectly on computer generated pdfs

I understand that there is music glyphs - https://www.notationcentral.com/product/musglyphs/
so I wonder if this glyph technique / or new research could be done to rebuild scanned pdfs into fresh pdf documents.
in this scenario - text is not so important - it would be image to image - but I need a base model that is suitably trained on music scores. Resolutions are a big problem though.