Is My Photo Too Noisy? Grain vs Detail

A photo shot at ISO 6400 looks grainy. So does a photo of a gravel path at ISO 100. One is a fault and one is the subject, and a check that cannot tell them apart is worse than no check.

Where the measurement is taken

Noise is estimated in the flattest areas of the frame — the tiles with the least variation in them. A clear sky, a painted wall, a shadow with nothing in it. In a flat area, anything that varies is noise by definition, because there is no detail there to confuse it with.

The number that comes out is a sigma in grey levels: how far a pixel typically sits from its neighbours in a region that should be uniform. Under about 1.5 grey levels is clean. Past 7 is heavy.

That is why the measurement is taken at native resolution and never on a downscaled copy. Downscaling averages neighbouring pixels together, which is exactly what noise reduction does — measure a shrunk image and every file looks clean.

When the check should say nothing

Run that method over real photographs and it produces confident wrong answers on a specific class of image: frames with no flat area anywhere. Dense foliage. An aerial shot of a tea plantation. A close-up of fabric.

In those, the flattest tiles available are still texture, and the check reports texture as noise. Two real photographs were called “heavy noise” for exactly this reason before it was fixed.

The fix is to measure how flat the chosen tiles actually are and decline when they are not flat enough. Measured across real photographs, the tiles sit at 2.6 on the median image and 12.5 at the 90th percentile, against 17.2 and 26.3 on the images that got false noise calls. A cut-off at 14 removes every false call and still measures 73 of 80 images.

So a missing noise reading on a photo of a hedge is the check working, not failing.

The reading that makes it useful

A sigma on its own tells you there is grain. It does not tell you whether that was avoidable.

The ISO is in the file’s own metadata, and it is worth reading only beside a noise finding. A grainy frame at ISO 6400 is a photograph taken in the dark and the grain is the price; a grainy frame at ISO 100 is something else — a heavy crop, an aggressive lift in post, or a sensor problem. Same measurement, completely different answer, and the difference is one field in the header.

A sharp ISO-6400 file gets no ISO note at all, because there is nothing to explain.

What to do about it

Noise reduction trades grain for detail and the trade is not reversible. Before reaching for it, check whether the grain is actually at a level that will show in the final medium: a print at arm’s length hides far more than a 100% crop on a screen, and a web image at 1600px wide hides more still.

If the file is already noisy and already exported, the honest answer is that the information was lost at capture. The fix lives at the next shoot — more light, a faster lens, or a lower ISO with a slower shutter if the subject allows it.