Magnification vs resolution vs accuracy

Magnification: how big it looks

Magnification is a ratio between the size of the image and the size of the object. Nothing more. It says how much of your screen a feature occupies.

It is also the least well-defined of the three, because on a digital system the honest answer depends on the monitor. The same camera and lens produce a different "magnification" on a 24-inch screen than on a 32-inch one, at no point having changed anything optical. This is why serious specifications quote field of view or micrometres per pixel instead.

Resolution: what can be told apart

Resolution is the smallest separation at which two features remain distinguishable as two features. It is limited by two independent things, and the worse one wins.

Optical resolution

Set by diffraction and by the numerical aperture of the objective. Light passing through a finite aperture cannot form an arbitrarily small point, and no amount of subsequent magnification recovers detail the optics never delivered.

Sensor resolution

Set by how finely the image is sampled. Two features that the optics resolve perfectly well are still lost if they both land on the same pixel.

A system with excellent optics and a coarse sensor is sensor-limited. A system with a superb sensor behind poor optics is optics-limited. Improving the wrong one changes nothing, which is why "we upgraded to a higher-megapixel camera and it measures exactly the same" is such a common and expensive discovery.

Accuracy: whether the number is right

Accuracy is how close your reported value is to the true value. It is the only one of the three that your customer, your auditor and your scrap rate care about, and it depends on far more than the optics:

• the calibration — its artefact, its span, its currency
• whether X and Y were calibrated separately
• lens distortion across the field of view
• how repeatably an edge can be located, which depends on focus, lighting and contrast
• whether the part is in the plane you calibrated in
• operator judgement in placing the endpoints

Two terms worth keeping separate: resolution is the smallest change you can detect; accuracy is whether the value is correct. A system that resolves 0.5 µm and carries a 3% scale error detects tiny changes about a number that is wrong. It is precise and inaccurate, which is the most dangerous combination because it looks convincing.

Empty magnification

Magnifying beyond what the optics resolve is called empty magnification. The image gets larger; the information does not increase. Edges become soft, wide and ambiguous, and the operator now has to decide where inside a fuzzy 8-pixel band the edge "really" is.

That decision is where the measurement is actually lost. Two operators will place it differently, and the same operator will place it differently twice. The system feels more precise because everything is bigger, and measures worse because nothing is sharper.

The practical consequence is counter-intuitive and worth stating plainly: zooming in further often makes a measurement less accurate, not more. The magnification that measures best is usually the lowest one at which the feature still spans a healthy number of well-defined pixels.

Choosing sensibly

1. Start from the tolerance, not the feature. A ±0.05 mm tolerance and a ±0.005 mm tolerance on the same feature are different instruments.
2. Work out the pixel scale you need. As a rule of thumb, you want the tolerance band to span a good number of pixels — not one or two — so that edge-placement ambiguity is small compared with the thing you are judging.
3. Check the whole feature still fits. A scale fine enough to resolve the tolerance is no use if the feature no longer fits in the frame. This trade-off is the real constraint in most applications.
4. Prefer telecentric optics for dimensional work where you can. They largely remove the magnification change with height that otherwise makes every reading depend on how the part is sitting.
5. Then light it properly. Backlighting a silhouette gives an edge that everyone agrees on. Front lighting a shiny machined face does not. Lighting routinely contributes more accuracy than the last increment of optics.

Measuret reports in the calibrated unit rather than in magnification, and stores calibrations by name so the configuration you chose is the configuration you measure in. More on calibration.