Notalo › Learn › How to photograph sheet music so a scanner can read it

How to photograph sheet music so a scanner can read it

By Taekyung Lee · 2026-10-10

Most conversions that go wrong on Notalo go wrong before the upload. The page was photographed at an angle, half of it in shadow, or so far away that a quarter note is eleven pixels tall. This is what we have learned from the pages people send us when a result looks bad, and what to do instead.

1. Shoot straight down, not from a chair

A page photographed from where you sit is a trapezoid: the top edge is shorter than the bottom, and the five staff lines, which should be parallel, converge. Our first processing step straightens the page by measuring the dominant line angle, but it corrects rotation, not perspective. A tilt of a few degrees is fine. A photo taken from 30 cm above the music stand, looking down at 45°, is not: the staff spacing changes from one side of the page to the other and the note detector, which expects every staff on the page to have roughly the same spacing, starts missing noteheads on the far side.

Put the page flat on a table, hold the phone directly above it, and look at the edges of the paper on screen. If they are parallel to the edges of the screen, you are done.

2. One light source, from the side, no flash

The flash on a phone is a point light right next to the lens. On glossy paper it makes a bright oval in the middle of the page where the ink disappears, and on matte paper it leaves the corners dark. Both are a problem because the second processing step turns the photo into pure black and white, and it has to pick a threshold that works across the whole page. A bright patch becomes white (notes vanish); a dark corner becomes black (staff lines merge).

Daylight from a window to one side is ideal. A desk lamp off to the side is nearly as good. Overhead room light is acceptable if you keep your own shadow off the page, which usually means standing so the light is in front of you, not behind.

3. Fill the frame with the page

Resolution is the quiet killer. A phone takes a 12-megapixel photo, but if the page is a third of the frame, the music is 4 megapixels, and a staff that is 8 mm tall on paper ends up about 60 pixels tall in the image. The detector was trained on scores where the space between staff lines is 12 to 20 pixels, and it degrades below about 10. In our test set, the same page photographed close (staff spacing 18 px) reads 99% of the pitches; photographed from twice the distance (9 px) it reads 86%, and the note lengths get worse faster than the pitches, because a dot after a note is only two or three pixels across.

Get close enough that the page fills the screen with a small margin. If a page has very small print, photograph the top half and the bottom half separately and upload them as two images.

4. Flatten the page

A book that will not stay open produces a curved page, and a curved page gives curved staff lines. We handle gentle curvature, but a strong curl near the spine does two things: the lines bend out of the horizontal window the staff finder looks in, and the noteheads near the spine are compressed sideways and look like something else. Hold the book down with a hand on each side outside the music, or photocopy the page first; a copy lies flat.

5. Avoid shadows from your hand or phone

The most common artifact we see is a soft dark band across the page from the phone's own shadow. It is wide enough to shift the black-and-white threshold inside it, which turns thin staff lines into thick ones. Stand so the light comes over your shoulder from the opposite side of the phone, or simply photograph in brighter light so the shadow is faint.

6. PDF beats photo, and a scan beats both

If the music came from a PDF, upload the PDF, not a screenshot of it. A PDF exported from notation software is rendered by us at 200 dots per inch into a perfectly aligned, perfectly lit image, and on those pages our detector finds every pitch on our test set. A scan from a flatbed scanner at 300 dpi is nearly as good. A photo is always a step down, which is fine for a clean page under good light, and a struggle for a dim, tilted one.

If the PDF is itself a scan of a printed book, it will behave like a scan, not like a rendered file. That is usually still good; the one thing that catches us is a scan that was made slightly rotated, which we straighten automatically.

7. Crop the page, not the music

Cropping away the table, the music stand and the neighbouring page helps, because anything with long horizontal lines (a table edge, a ruled notebook) can be mistaken for a staff. Cropping into the music hurts: a staff that is cut off at the left edge loses its clef, and without a clef we assume treble, which gives the left hand the wrong letters by a sixth. Leave a finger's width of white around the music.

What we do with the photo

For the curious: the image is converted to greyscale, the skew angle is measured from the staff lines and corrected, uneven lighting is removed by subtracting a blurred copy of the page from itself, and the result is thresholded to black and white. The staff lines are found from runs of black pixels of consistent length and spacing. Then a neural network looks at the page and marks noteheads, clefs, sharps, flats and naturals. The rest is bookkeeping: which staff each notehead is on, how far from the bottom line, which clef and key signature apply, and whether an accidental earlier in the bar changes it.

Every step before the neural network is there to make a photo look as much like a rendered PDF as possible. The better the photo already looks, the less those steps have to do, and the fewer mistakes they make. That is the whole reason for this list.

A quick checklist

• Page flat, phone directly above, paper edges parallel to the screen.
• Light from a window or lamp to the side; flash off; no shadow across the page.
• Page fills the frame. Small print: two photos, top and bottom.
• Book held open or photocopied; no curl at the spine.
• Leave white margin around the music; crop away everything else.
• If you have the PDF, send the PDF.

Written by Taekyung Lee, a self-taught pianist (four years, no lessons) who built Notalo for personal practice. About · Instagram

More: all guides and articles · demo songs with letters.