What optical music recognition must recover
Text OCR reads characters in a line. Music notation is two-dimensional: the vertical position of a note determines pitch, horizontal grouping affects rhythm, and symbols can apply to one note, one voice, one staff, or an entire system.
An OMR system must locate staff systems, remove perspective distortion, recognize symbols, infer relationships, and encode the result in a machine-readable score. Research surveys consistently identify complex layouts, degraded pages, handwriting, and inconsistent evaluation as major challenges.
How to photograph a score for better recognition
Recognition cannot restore information that is blurred, clipped, hidden by glare, or compressed into too few pixels. Improve the source before uploading it.
- Place the page flat and photograph it directly from above.
- Use even light and avoid shadows across staff lines or glossy reflections.
- Include the full page while keeping the notation large enough to resolve ledger lines and dots.
- Remove perspective skew and rotate the page so staff lines are horizontal.
- Scan one page at a time and keep page order outside the recognition step.
- Prefer a clean original or high-resolution PDF over a screenshot of a compressed image.
Review the musical relationships, not only the symbols
The most damaging errors are structural: a missing voice, incorrect duration, wrong measure boundary, or accidental attached to the wrong note. Compare each measure with the source and listen to playback before adjusting layout.
Check clefs, key and time signatures, tuplets, ties, dotted rhythms, cross-staff notation, lyrics, and repeat structures. Handwritten annotations and unusual contemporary notation may need manual re-entry even when the printed notes are recognized well.
Choose an editable output
PDF and PNG preserve appearance but do not become editable music. MusicXML is the practical output when you want to correct notation in another editor, while MIDI is useful when you mainly need pitch and performance events.
NotationAI turns the recognition result into the same score used for editing, playback, practice, and collaboration. Keep the original page visible during verification and preserve it with the project so future editors can resolve uncertain passages.
Continue with an editable score
Capture, edit, arrange, practice, and collaborate in the same NotationAI product across App and Web.
Frequently asked questions
Can a photo of handwritten music become an editable score?
Sometimes, but handwriting varies far more than printed notation. Expect more correction, especially for rhythm, slurs, voices, and nonstandard symbols.
Why is MusicXML better than PDF after scanning?
PDF describes a page image or layout. MusicXML describes musical structure that notation software can edit, play, and reformat.
Does a higher-resolution photo always improve OMR?
Only until symbols are clearly resolved. Focus, lighting, perspective, cropping, and page condition can matter more than adding more pixels to a poor capture.