Monochromatic lighting in machine vision: why one wavelength wins

Two inspections through the same lens can resolve different detail: one under white light, one under a green LED, and the green image holds sharp edges where the white one smears. The cause is chromatic aberration. A lens that must perform from the blue to the red end of the visible band spends much of its design budget holding focus across it, and the residual colour error often sets the spot size before diffraction does. Light the part with a single wavelength and that term drops out of the budget. This explainer gives the numbers: the diffraction floor at each wavelength and f-number, a worked example at 200 mm working distance, the design-band rule for your lens, and the choice between a monochrome LED light and a bandpass filter on the white light you already own.

What does one wavelength change inside the lens?

It deletes chromatic aberration. Refractive index falls as wavelength rises — about 0.008 index units from the blue to the red line on common crown glass — so no single-element shape brings all colours to one focus. One wavelength removes the term instead of paying for glasses and elements to cancel it.

The lens maker's arithmetic:

Lens power scales with (n − 1), so a single element's focal length shifts by 0.008 / 0.52 ≈ 1.5% across the visible band — about 0.8 mm at a 50 mm focal length. Multi-element FA lenses add doublets and special glasses that cancel most of that shift, but the residual colour error is never zero, and stopping down does not fix it: small apertures cure geometric aberrations, not colour.

The converse is good news. A broadband-corrected lens works perfectly well under monochromatic light, as long as the wavelength sits inside its design band — the check comes in the design-band section below. How much of a given lens's budget went into colour correction shows in its data: the lens performance curves guide reads MTF against aperture and field.

What resolution does the diffraction limit set at a single wavelength?

Diffraction sets the floor: the Airy diameter is about 2.44 × λ × f-number, so at 525 nm and f/2.8 no lens images a point smaller than roughly 3.6 µm, and at 660 nm the same aperture gives about 4.5 µm. In white light, colour blur sits on top of this floor.

λ is the wavelength, N the f-number. At f/2.8 the diameter works out to 2.44 × 0.450 µm × 2.8 ≈ 3.1 µm in blue, 3.6 µm in green, 4.5 µm in red. Stop the same lens to f/8 in green and it grows to 2.44 × 0.525 µm × 8 ≈ 10.3 µm. These are ideal-aperture floors; a real lens adds geometric aberration on top, and only white light adds colour — which is why a single wavelength is what lets an ordinary FA lens actually reach its diffraction number. What the number means for discernible detail is the subject of the lens contrast and resolution guide.

How much sharper will the image actually get?

Gains show once the diffraction spot alone is near or below your pixel pitch: at f/2.8–f/4 that means pixels of 2–3.5 µm. When the Airy diameter already covers the pixel (f/8 on 4.5 µm pixels), the aperture was your limit and monochromatic light buys contrast, not resolution.

Worked example. The camera is a 20 MP 1.1" sensor with 2.74 µm pixels (4504 × 4480), the working distance 200 mm, the light a 525 nm LED pulsed at 100 µs, the part moving at 0.5 m/s. The lens is the LS-FFL-0567, a 16 mm C-mount lens rated 20 MP on 1.1" with F2.8–22, set to f/2.8.

1. Sensor width: 4504 × 2.74 µm ≈ 12.3 mm.
2. Field of view: FOV = 12.3 mm × 200 mm / 16 mm ≈ 154 mm, so the object-space pixel is 154 mm / 4504 ≈ 34 µm.
3. Diffraction at 525 nm, f/2.8: d = 2.44 × 0.525 µm × 2.8 ≈ 3.6 µm = 1.3 pixels, which is 34 µm × 1.3 ≈ 45 µm in object space.
4. The same check at f/8: 10.3 µm = 3.7 pixels ≈ 130 µm in object space — stopping down costs three times the resolution, and the colour term stops mattering next to it.
5. Motion: 0.5 m/s × 100 µs = 50 µm of blur, about 1.5 pixels; at 500 µs it would be 250 µm, around five times the optical floor.

At f/2.8 in green, the pixel (34 µm) and the diffraction floor (45 µm) sit within half a pixel of each other, so holding the colour error at zero is what keeps this station near 45 µm in object space. The general form, via magnification, is in the object-space pixel size guide.

Does any wavelength work, or must it fall inside the lens's design band?

It must sit inside the band the lens was designed and coated for. A visible-band FA lens behaves well under 450–680 nm LEDs; drive it at 940 nm and focus shifts while coating transmission falls, so the fix is an IR-corrected lens, not more light.

Two things break outside the band. Focus moves, because the correction balances the design wavelengths rather than yours — the image stays bright but lands off best focus. And transmission drops, because the anti-reflection coating is tuned to the design band; past it, reflections and veiling glare climb. The check costs one minute: refocus at the working wavelength and see whether the sharpness you had in white comes back.

A standard 35 mm, 2/3", 6 MP C-mount FA lens such as the LS-FFL-0679 is corrected and coated for roughly the visible band. For work beyond 850 nm, choose the shop's IR-corrected lens type — 16 models — and refocus at the working distance. Where the interesting band sits, physically, is mapped in the wavelength ranges guide and the SWIR vs NIR comparison.

Is a bandpass filter on white light better than a monochrome LED light?

Optically the two are the same: once one narrow band reaches the sensor, chromatic aberration is gone either way. The filter is the less expensive retrofit but costs light — a 10 nm bandpass over a 400–700 nm source passes roughly 3% of the power — so expect longer exposure or more gain.

The filter also blocks ambient light of every other colour — the same job a strobed monochrome LED does in a dark hood. Filter families and where each pays off are compared in the optical filters guide.

Which controller settings keep wavelength and intensity stable?

Drive the LED from a constant-current output and strobe it. Constant current pins the junction current, so intensity stays fixed and wavelength drifts only with junction temperature — roughly 0.03–0.05 nm/K for green, under 2 nm across a 40 K swing, well inside any 10 nm filter or the lens's band.

Short pulses help twice over: they freeze motion at the 100 µs scale of the example above, and they keep the junction cool, which is what pins the wavelength. Timing has its own trap — switching delay eats microseconds from both ends of the pulse — covered in the LED strobe rise time guide.

Three controller types cover the usual sizes: a 4-channel, 120 W constant-current unit for 24 V lights; a 4-channel, 200 W digital box with strobe; and a high-power constant-current unit with 240 W per channel for large strobed arrays.

When is monochromatic lighting the wrong choice?

When the detail you need lives in colour. One wavelength turns a multicolour part into a bright-or-dark map — matched colours go bright, everything else goes dark — so colour grading, print inspection and colour-presence checks stay on white light with a broadband-corrected lens.

It is also the wrong tool when the aperture already sets the limit: at f/8 the diffraction diameter alone is 10.3 µm in green, more than twice a 4.5 µm pixel, so colour correction was never what held the image back. And think twice on a colour camera: a Bayer array wastes most of a monochrome light, as the FAQ below explains.

If the part is colourless in inspection terms — metal, glass, most mouldings — pick the wavelength by contrast rather than by diffraction. The LED colour contrast guide and the monochrome camera contrast guide score the choices, and sequencing colours through one camera can even replace white light.

Which combination should I order?

Pick by pixel pitch and by what the part carries. A green LED light with a corrected 16 mm-class lens for 2–3.5 µm pixels run at f/2.8–4; a 10 nm bandpass on the white light you already own when the pixels are that small and exposure allows about 30× more; white light and a broadband lens whenever the part itself is coloured.

The numbers, from the example above: at f/2.8 in green the station floor is about 45 µm in object space; at f/8 it is about 130 µm whatever the light; a colour part re-routes the decision to the lens band, not the wavelength. If your numbers differ, the same three checks still settle it: Airy diameter versus pixel, motion blur versus pulse length, wavelength versus lens band.

Browse machine vision lenses and light controllers, size the drive in the light controller finder, or start from the part in the lens finder. The contact page takes a drawing review.