What Limits Online PD Accuracy? The Five Sources of Error

Last updated: October 2026 | Reading time: 9 minutes

Short answer: an online PD measurement is limited by five things: how precisely the pupil centres are located, how well the scale object is detected, the depth between that object and your eyes, how far your head is turned, and how your eyes converge on the camera. Together they decide whether you are within a millimetre or two of the true value.

A 1 mm error on a 63 mm PD is about 1.6%. That is small, which is exactly why each of the factors below matters: none of them can be larger than a millimetre or so if the result is to be trusted.

1. Locating the pupil centre

Software finds the iris with a face-landmark model and, in our tool, refines the horizontal centre of each iris by fitting a circle to the edge between iris and white of the eye. On a typical phone photo, one pixel can be a fraction of a millimetre, but eyelids covering the top of the iris, glare, and low contrast between the iris and the white of the eye all make the edge harder to find. The refinement uses only the left and right arcs of the iris, because the eyelids hide the top and bottom.

What you can do: use even light, open your eyes fully, and take several photos.

2. Detecting the scale object

A bank card is 85.60 mm wide. If the software mis-measures its width by 1.5 pixels on a 400-pixel card, that is a 0.4% error. A tilted or half-hidden card can do far worse.

What you can do: keep the card flat with all four corners visible.

3. Depth between the card and the pupils

This is the largest source of systematic error. A card on your forehead sits closer to the camera than your pupils. An object that is closer looks bigger, so without a correction the pupils appear farther apart than they are in millimetre terms. The size of the correction depends on how far behind the card your pupils sit, which varies from person to person, and on your distance from the camera.

Our tool uses an assumed average offset. In our simulations, if a person's real offset differs from the average by 5 mm the result moves by roughly 0.8 mm; if it differs by 10 mm, the error can exceed a millimetre. This is why we plan a one-time personal calibration against a trusted PD and why we do not publish an accuracy percentage yet.

4. Head turn

A turn of the head by an angle θ shrinks the horizontal gap between your pupils in the image by about cos θ. At 5° that is 0.4%; at 12° it is about 2%. The tool therefore rejects photos with a head turn above about 6° and corrects for the remainder.

5. Convergence

Looking at something close makes your eyes turn inward. At camera distances of 35 to 50 cm, the pupils end up roughly 1.2 to 1.7 mm closer together than when you look into the distance, based on the geometry our tool uses. If you want your distance PD and you photograph yourself at arm's length, this inward turn must be added back. The tool does this automatically; many simple apps do not.

What the independent evidence says

A published study compared three smartphone PD apps with a digital pupilometer in 44 participants and found that the apps differed in error (PMC10389117). The spread between apps is itself a reminder that the design of the method matters more than the fact that it runs on a phone.

How the tool reports uncertainty

Instead of a vague "95% confident" label, the result shows an estimated error in millimetres, computed from the factors above. The estimate is model-based: it describes how the tool's own method behaves, not a comparison against clinic measurements. We say so openly because honest limits are more useful than a bold claim.

When to see an optician

• Progressive or bifocal lenses
• Strong prescriptions
• Glasses for children
• Any time a number surprises you

See PD for progressive lenses and PD for high prescriptions.

Related reading

How to measure PD with a credit card · Online PD apps: how accurate? · Near PD vs distance PD

Sources

• MediaPipe Iris, Google Research: iris-based depth estimation and its accuracy claims.
• Comparing the effectiveness of smartphone applications in the measurement of interpupillary distance (PMC10389117).