Lens coating and spectral transmission: why contrast drops

A machine vision lens that resolves beautifully on the bench can still hand you a flat, hazy image once you move to an 850 nm backlight or a 365 nm UV spot. The cause is rarely the sensor. It is what the lens glass and its coatings do at your wavelength, and it is predictable arithmetic, not guesswork. This explainer walks through surface reflection, what an anti-reflection coating buys you, where the losses turn into contrast-killing stray light, and why the spectral transmission curve rather than the f-number decides whether a lens works in the near infrared or the ultraviolet. You will be able to estimate the loss for any element count, convert an f-number into a T-stop, and choose between standard, NIR-coated, SWIR and quartz UV optics with numbers.

How much light does an uncoated lens really lose?

An uncoated glass surface reflects about 4% of the light arriving on it, and the losses multiply: a six-element lens with twelve glass-to-air surfaces passes only around 61% of the incoming light and scatters the remaining 39%.

The reflection comes from the refractive index step at each glass-to-air boundary. At normal incidence the Fresnel equations give the reflectance R = ((n − 1)/(n + 1))². For a typical crown glass with n = 1.5 that is (0.5/2.5)² = 0.04, so 4.0% of the light bounces off every surface. Higher-index flints near n = 1.6 reflect (0.6/2.6)² ≈ 5.3%, which is where the familiar 4-5% figure comes from. Absorption inside good optical glass is small across the visible band, so reflection dominates the budget.

With N surfaces the transmission multiplies rather than adds:

For the six-element example: 0.96^2 = 0.92, 0.96^4 = 0.85, 0.96^8 = 0.72, and 0.96^12 = 0.72 × 0.85 = 0.61, i.e. 61%.

The missing 39% does not vanish. Nearly all of it re-enters the barrel as scattered light and ghosts. A protective window in front of the lens adds two more surfaces at the same rate: 0.61 × 0.96^2 = 0.56, so the same lens behind a window passes about 56%. What a window otherwise changes at your wavelength is covered in what a protective filter window changes.

What does the anti-reflection coating on a lens actually do?

The green or reddish shimmer is a stack of quarter-wave interference layers, each about 100 nm thick, deposited so the reflected beams cancel each other; a good multilayer stack holds reflection below about 0.5% per surface across its design band.

Each layer is an odd number of quarter-wavelengths thick in optical path. Light reflected from the top and bottom of a layer arrives half a wavelength out of phase and cancels by destructive interference, while the transmitted beams add. A single quarter-wave magnesium-fluoride layer tuned for 550 nm has a physical thickness of 550/(4 × 1.38) ≈ 100 nm, and it drops that one surface from 4.0% reflection to about 1.4%. Stacks of many oxide layers hold residual reflection to roughly 0.2-0.5% per surface across the design band, which is how a well-coated six-element lens reaches the ~94% total transmission used in the next section.

That stack is also a real cost driver: many layers, tight thickness control and high-vacuum deposition separate a basic lens from a quality one. The band matters as much as the residual number. A coating tuned for 400-700 nm is at its best inside that range and gives performance back outside it. The same interference physics, pushed further, builds the band-pass filters in colored glass vs interference filters.

Why does stray light from lens surfaces lower image contrast?

Every reflected percent can return as veiling glare spread over the whole image, and 15% stray light is enough to turn a 10:1 feature contrast into 4.6:1 — the difference between a clean threshold and a noisy one.

Reflected light is not simply lost. Most of it re-enters the image path after further bounces and arrives as a diffuse veil plus discrete ghost images, and the veil raises every dark pixel toward the bright level:

• Bright background images at 200 DN, dark defect at 20 DN: object contrast 200/20 = 10:1.
• Stray light at 15% of the bright level adds L = 0.15 × 200 = 30 DN to every pixel.
• Measured contrast: (200 + 30)/(20 + 30) = 230/50 = 4.6:1.

If your threshold logic needs 6:1, the lens fails the inspection even though the sensor saw the defect clearly. The 15% is an example assumption; measured veiling glare varies lens to lens and grows fast with bright sources in or near the field. The stray-light diagnosis guide turns that symptom into a cause, and lens contrast and resolution covers the other half of contrast, the modulation the optics deliver at your feature size. Small pixels suffer twice — less light per pixel and tighter thresholds — which is why small-pixel camera selection weighs contrast as heavily as resolution.

How much exposure time do transmission losses cost you?

Coating losses act like a slower aperture: at F2.8 with 85% total transmission the lens behaves like T3.0, about a quarter of a stop slower, and an uncoated six-element design at 61% runs roughly three-quarters of a stop behind.

Photography solves this with the T-number, T = f-number/√τ, where τ is the lens's total spectral transmission:

Exposure scales by the same factor: an exposure of 2.0 ms that just fills the sensor through an ideal F2.8 needs 2.0/0.61 = 3.3 ms through the uncoated six-element design, and 2.0/0.94 = 2.1 ms through the coated one. If your line rate allows 3.0 ms, the lens decides pass and fail while nothing else in the system changed. One current FA lens of the class these numbers describe:

The LS-FFL-0567 is a 16 mm, F2.8-22 C-mount lens for 1.1" sensors up to 20 MP, with distortion below 0.3%. Transmission τ is roughly a constant multiplier as you stop a given lens down, so the penalty is a fixed share of your light budget — worst when light is scarce. Stopping down also trades light for aberration correction; that trade is covered in how stopping down helps aberrations, and the exposure-versus-gain interaction in automatic iris, exposure and gain control.

Why does a visible-light coating fail in the near infrared?

A standard lens is corrected and coated for roughly 400-700 nm, and its coating reflectance climbs again beyond that band, so at 850-950 nm you get rising stray light exactly where NIR illumination puts all its energy.

The glass is not the bottleneck: common optical glasses transmit well past 1 µm. The coating is. A stack tuned for 400-700 nm leaves its low-reflection trough behind, and by 850-950 nm the per-surface reflectance climbs again, so a standard FA lens under an 850 or 940 nm ring light returns a hazy, low-contrast image for exactly the reason computed above. Machine-vision practice is to insist on glass and coatings specified for the band. For choosing the band itself, see SWIR vs NIR vs visible imaging and near-infrared wavelength selection for water and silicon.

Lenses specified and coated for the short-wave infrared band:

The LS-SWL-0246 covers 400-1700 nm at F1.8 on a 1/2" sensor, the LS-SWL-0229 gives 35 mm and F1.4 for 1" sensors across 900-1700 nm, and the LS-SWL-0256 reaches 100 mm at F2.1 for 4/3" sensors with an F-mount. All three sit in the thermal, SWIR and UV lens category. If you run a color sensor, its IR-cut filter blocks most of the NIR anyway — color camera NIR imaging and IR-cut options explains why NIR work usually lands on monochrome sensors.

Why can't a standard glass lens see below 350 nm in the UV?

Crown and flint glasses absorb ultraviolet light: transmission falls steeply from about 350 nm and is under 5% near 310 nm, so a standard FA lens is effectively blind there, while fused quartz keeps transmitting down to about 180 nm.

Below the visible band the failure mode flips from reflection to absorption. Crown and flint glasses absorb UV inside the material, so no coating can recover it — the loss is not happening at a surface. The practical reading: 365 nm inspection with a standard lens is already marginal, and 310 nm or below needs UV optics. Fused quartz is harder to source and to figure, which is why UV lenses exist as dedicated designs rather than options on a visible FA lens. For the imaging techniques, see SWIR and UV lens basics and reflected UV vs UV fluorescence.

Purpose-built UV optics in the same category:

The LS-SWL-0291 is a 25 mm, F2.0 C-mount lens specified 240-1000 nm for 2/3" sensors; the LS-SWL-0280 covers 190-700 nm at a fixed 1.4x magnification with a C-mount.

When is the coating not the reason your image looks flat?

If contrast is low under visible light with a modern multilayer-coated lens, suspect the lighting geometry first: a wrong illumination angle can swallow all the contrast at once, while coating losses are a fixed percentage the light budget was sized for.

Run one test before blaming the optics. Image a matte black patch inside an evenly lit white field, exposed so the white sits near 200 DN. If the black patch reads tens of DN, surface glare or a light source shining straight into the front element is feeding the sensor; fix it with baffling, a repositioned light or a better-coated lens. If black sits near zero but the part still shows weak contrast, the cause is geometry or wavelength: work through lighting geometry for image contrast and which LED colour gives the best contrast. Lens type is its own decision — entocentric vs telecentric lenses.

How do you choose a lens for your wavelength band?

Match the coating band to the illumination band first and the glass second: confirm the transmission curve covers your LED wavelength inside its low-reflection band, then compare f-number, distortion and resolution.

Checks before ordering: get the relative spectral transmission curve and read it at your LED wavelength, not as a band average; run the black-patch glare test at your working f-number; convert f-numbers to T-stops when comparing lenses for a tight exposure budget. Filter by sensor format and focal length in the machine vision lenses category or the lens finder, and work camera-to-lens step by step in the lens configurator. If the deciding document is a transmission curve for your band, ask for it through the contact page — we supply it with the quotation. More on matching optics to sensors sits on the lenses topic page.