Imaging Glossary › Nyquist Ratio
Sampling Metric

Nyquist Ratio

Are you sampling enough?

View
Definition

The ratio of optical resolution to sampling interval, indicating whether the pixel size properly captures the detail the objective can resolve. A ratio of 2.0-2.5 is optimal; below 2.0 means undersampling (losing resolution), above 3.0 means oversampling (wasting pixels).

Pixels per resolution element

The Nyquist ratio is how many pixels span one resolution element: ratio = resolution limit ÷ pixel footprint at the specimen, where the footprint is pixel pitch ÷ total magnification. Nyquist's requirement of two samples per cycle is met at a ratio of 2. Below 2 the grid is too coarse for the detail the optics deliver and aliasing follows; above about 3 the same photons are divided among more pixels without recording anything new.

A note on a discrepancy you may meet elsewhere, including in an earlier version of this entry: some write the ratio with a factor of two in the denominator, which shifts the satisfied condition to 1 rather than 2. Both conventions describe the same physics; mixing them produces thresholds that disagree with the formula beside them. The pixels-per-resolution-element form is used throughout this glossary because it can be read directly as a count.

Simplified

The number answers one question: how many pixels does the smallest detail your lens can deliver land on? Fewer than two and the detail is lost, and worse, replaced with a false pattern. More than about three and you are splitting the same light across extra pixels for no new information. Two to three is the range where the camera and the objective are telling each other the truth.

Reading the number, and why 2 to 3 rather than exactly 2

Two is the theoretical floor, and practice sits a little above it. The reason is contrast rather than geometry: a lens does not transmit detail at full contrast right up to its cutoff, it fades toward it, and a pixel integrates over its own area, adding a second attenuation that grows as the footprint approaches the limit. Sampling at 2.3 to 2.5 keeps most of the transferred contrast rather than only the bare information, and it gives deconvolution something to work with.

Beyond about 3 the returns reverse. The same photons spread over more pixels, so each pixel's signal falls while read noise is paid once per pixel regardless — a real cost at very low light, and a reason oversampling is cheap rather than free. The saving grace is that it is reversible: sum the pixels afterwards and you have exactly the coarser image. Undersampling offers no equivalent undo.

Simplified

Two pixels per detail is the mathematical minimum; a little over two is the practical sweet spot, because fine detail arrives faint rather than crisp and a slightly finer grid holds on to more of it. Go far past three and you are dividing the same light among more and more pixels, each of which costs you a little noise to read. The consolation is that you can always add oversampled pixels back together later. You can never undo pixels that were too big.

Two systems, worked through

A 3.7 µm sensor at 25x, 1.1 NA, 525 nm. Footprint = 3700/25 = 148 nm. Resolution = 0.61 × 525 / 1.1 = 291 nm. Ratio = 291/148 = 1.97 — a hair under the floor of 2, so this pairing is at the boundary: essentially all of the optical detail is captured, with no margin for the contrast loss near the cutoff. Sound in practice, and not a combination to bin.

A 6.5 µm sensor at 40x, 1.25 NA, 560 nm. Footprint = 6500/40 = 162.5 nm. Resolution = 273 nm. Ratio = 273/162.5 = 1.68 — undersampled, losing the finest detail the objective delivers. Moving the same camera to 60x gives a footprint of 108 nm and a ratio of 2.52, comfortably in range. Nothing about the camera changed; the pairing did.

Simplified

Run the number for a real pairing and it answers immediately. A 3.7 micrometer sensor behind a 25x lens gives 1.97 — right at the edge, capturing what the lens delivers with nothing to spare. A 6.5 micrometer sensor behind a 40x lens gives 1.68, which means detail is being lost before it is recorded. Put that same camera behind a 60x lens and the number becomes 2.52, which is comfortable. The camera was never the problem; the pairing was.

Connected Terms

Share This Term
Term Connections