Camera sensitivity vs responsivity: why DN per noise matters more

A camera that turns the same light into more DN reads like the better camera, and a spec sheet rewards that impression. Responsivity — DN per µJ/cm² of sensor exposure — is easy to inflate with gain, and the noise rises with it. The number that actually ranks two cameras is signal-to-noise at the exposure your scene delivers. Here you get the response-curve vocabulary (NEE, SEE, responsivity, peak-to-peak noise), a worked two-camera comparison with the arithmetic on the table, and a shortlist procedure for ranking area scan cameras by signal-to-noise instead of by a headline DN figure.

The response curve: NEE, SEE and responsivity

Plot a sensor's output against exposure and you get three regions. In the dark region the output is low, noisy and unpredictable, so no usable measurement lives there. From the noise equivalent exposure (NEE) onward the output rises linearly: equal steps of light produce equal steps of DN. At the saturation equivalent exposure (SEE) the curve bends over, more light stops buying more signal, and bright features clip.

Classical sensitivity is the slope of that linear region, quoted in volts per µJ/cm² for an analog video signal. A digital camera outputs digital numbers (DN), so the honest name for the same slope is responsivity, in DN per µJ/cm² at a stated gain, wavelength and exposure time. Responsivity says how fast light becomes levels. It says nothing about whether those levels mean anything; that is noise's job, and noise decides every ranking below.

Gain manufactures responsivity, and the noise comes with it

Raise the gain by a factor k and every DN leaving the converter — signal and noise alike — is k times larger. The sensor collected exactly the same photoelectrons, while the DN per µJ/cm² figure went up by k. A gain setting can double or quadruple a responsivity number without changing what the camera can see: the photon budget is fixed by the sensor and the light, and gain only renames the units.

Responsivity is also spectral. One µJ of 850 nm light carries about 4.3×10¹² photons; one µJ at 460 nm carries about 2.3×10¹², because photon energy is hc/λ. The same sensor therefore prints different DN per µJ/cm² under red light than under blue, so put both cameras under your illumination wavelength before believing either figure.

The numbers that rank cameras: noise, NEE and dynamic range

Three measured numbers do the ranking. Noise, read as peak-to-peak for a quick check or as RMS temporal dark noise for a rigorous one, sets the smallest DN difference that means anything. NEE, the noise floor divided by responsivity, is the dimmest exposure that still produces a predictable output — the honest sensitivity figure, and invariant to gain, because raising gain multiplies n and R by the same k. SEE, or saturation capacity in electrons, is the brightest exposure before the response bends, and it decides whether highlights survive.

Dynamic range ties the two ends together:

A sensor with a 10,000 e⁻ saturation capacity and 5 e⁻ temporal dark noise has DR = 10,000 / 5 = 2,000, and 20 × log10(2,000) = 20 × 3.301 = 66 dB.

EMVA 1288, the machine-vision standard for measuring and presenting camera specifications, exists so this comparison is fair: it fixes the test conditions, and its parameters are exactly the quantities above — temporal dark noise, saturation capacity, dynamic range, SNR at saturation and quantum efficiency versus wavelength — reported so two documents from different manufacturers can be read side by side.

Worked example: two cameras at 4 µJ/cm²

The inputs are illustrative, but in the shape spec sheets actually arrive in. Camera one: responsivity 1 DN/µJ/cm², noise 2 DN peak-to-peak. Camera two: responsivity 2 DN/µJ/cm², noise 5 DN peak-to-peak, quoted at the same gain. The scene is a dim inspection point delivering H = 4 µJ/cm², above both noise floors and inside both linear regions. The rules: SNR = R × H / n, and a gain of k multiplies R and n by k alike.

Camera one at unity gain: signal = 1 DN/µJ/cm² × 4 µJ/cm² = 4 DN, so SNR = 4 / 2 = 2. Camera two: signal = 2 × 4 = 8 DN, so SNR = 8 / 5 = 1.6. Camera two looks twice as sensitive on paper; at the actual exposure it delivers 1.6 units of signal per unit of noise against camera one's 2.0.

Now double camera one's gain so the responsivity figures match. R becomes 2 DN/µJ/cm², noise doubles to 2 × 2 = 4 DN peak-to-peak, signal becomes 8 DN, and SNR = 8 / 4 = 2 — unchanged, as it must be, because gain adds no photons.

Decision. The spec sheets now read identically, 2 DN/µJ/cm² each, and camera one wins: SNR 2.0 against 1.6 at the same exposure is 25% more signal per unit of noise, and its floor sits at 2 µJ/cm² against 2.5 µJ/cm², so it also resolves dimmer features. The decision flips only if camera two's noise falls to 4 DN peak-to-peak (SNR = 8 / 4 = 2, a tie) or below it (then camera two wins). A real camera adds a little read noise as gain rises and shot noise grows with signal, so doubling gain rarely doubles noise exactly; the clean numbers are there to make the ranking logic visible.

When the higher-responsivity camera is the right pick

Gain-matched responsivity is a phantom, but native responsivity at a stated gain is real, and there it can decide the purchase. Higher R at equal noise means a lower NEE, which wins when the scene is dim and cannot be brightened: short pulses, fast conveyor lines, dark or poorly reflecting parts. Higher native responsivity also reaches your target DN with less gain, so the noise you compared stays the noise you get. Run the check at the other end too: a strong responsivity can saturate early and clip a mixed scene, so put your brightest region against the saturation exposure before committing. Once both cameras clear NEE and SEE at your exposure, SNR ranks them — never DN per µJ/cm² alone.

Checks before you order

1. Fix the exposure budget first: illumination, f-number and exposure time set H at the sensor, and every comparison below happens at that one number.
2. Pull noise and responsivity for each finalist at one stated gain, or take the EMVA 1288 parameters, and compare at your illumination wavelength.
3. Compute SNR = R × H / n at your H for each camera, then check NEE sits below your darkest feature and SEE clears your brightest.
4. Buy photons before gain: more light, a longer exposure or a wider aperture raises SNR permanently, while gain only rescales DN. The measurement-side version of this argument is worked through in brightfield noise: exposure before gain.
5. Measure noise yourself when reports are missing: cap the lens, record a series of dark frames at your gain, and read the frame-to-frame spread. The dark frame subtraction method shows how many frames it takes to separate fixed pattern from temporal noise.
6. Confirm the camera actually ran the gain and exposure you set; per-frame metadata is the receipt, and reading actual exposure per frame covers the case where the property reports unsupported.

A mono shortlist to start from

When color carries no information, mono is the better photon deal: each color photosite's Bayer filter passes only its own band and turns the rest of the spectrum away, while a mono sensor keeps it. Start the shortlist on mono sensors and spend the comparison effort on noise. Two mono formats to start the ranking from:

The camera finder applies the same filters — sensor format, resolution, interface — across the full range, and the camera step of the configurator assembles the lens and light side of the budget once the sensor is chosen.

The decision, case by case

• Same gain on both spec sheets: lower noise wins, and NEE gives the dim-light score.
• Different gains quoted: scale each noise figure by the gain ratio first; the headline DN per µJ/cm² has ranked nothing on its own.
• Dim scene you cannot brighten: rank by native responsivity and NEE, then confirm SEE stays out of the way.
• Bright, mixed scene: check saturation first, because a clipped highlight is a lost measurement.
• Every case: photons beat gain — light, exposure time and aperture first, gain second, bit depth last. Where auto exposure fits into that, manual exposure versus auto exposure works the comparison through.

Shortlist with those rules. The cameras topic page collects the surrounding arithmetic, and if you want a second pair of eyes on the ranking, send the exposure budget through the contact page and we will run the SNR comparison with you.