White surface texture inspection lighting: kill the glints

You are trying to inspect the micro texture of a white surface - something close to sandpaper - and in near dark field the image fills with hard specular glints. The instinct is to order a coloured ring light and let the wavelength do the work. It will not. Wavelength selects what a surface absorbs, and a white surface absorbs almost nothing; the glints are a geometry problem, and geometry is solved with angles or with polarization. This guide gives you the reflection condition that sets the illumination angle, the working-distance reality of dark-field ring lights, the crossed-polarizer route for truly random texture, and a complete camera, lens and light worked example with the arithmetic shown. You will leave knowing which angle class to buy, how close to mount it, and when to reach for polarizer film instead.

Why doesn't a coloured ring light fix the glints?

Because a facet that mirrors light into your lens does so at every visible wavelength almost equally: for a glossy polymer with a refractive index near 1.5, the surface reflection is about 4% at normal incidence and about 9% at 60°, and it barely moves from blue to red. The glint you see is that Fresnel reflection off the crests of the texture. Swap the white LED for red, blue or green and the glint and the white diffuse background change together, so their ratio - your contrast - stays exactly where it was. The refractive index of a polymer shifts only a little across the visible band, which is why the reflection follows it so weakly.

Colour earns its keep only where absorption differs: ink printed on the part, an oxide tint, a stain. Against white polymer it has nothing to grab. Which LED colour gives the best contrast in machine vision works through the cases where a wavelength genuinely changes an image; a white matte texture is not one of them.

Two levers actually remove a glint. Take the ray path away, so the mirror direction misses the lens - that is dark-field geometry. Or take the polarization state away, so the specular component cannot pass the analyzer - that is crossed polarization. The rest of this guide is those two levers, with numbers.

What angle keeps the specular reflection out of the camera?

Keep the incidence angle beyond twice the steepest facet slope plus the lens acceptance half-angle: at 35° facets and a 0.9° lens cone you need more than 71° from the surface normal, meaning under 19° of elevation - the low-angle (0-15°) class of ring lights. Aim the camera straight down the surface normal. A facet tilted by δ from the mean surface sends an incident ray arriving at incidence i (measured from the normal) into the lens whenever 2δ - i falls inside the acceptance cone of half-angle θa. Every illumination angle therefore has a victim slope, δ ≈ i/2. The image is clean only when the texture holds no facet that steep:

Stopping down helps only marginally: at m = 0.21, f/5.6 gives θa = 0.9° and f/8 gives 0.6°. The real lever is incidence. With θa = 0.9°, the decision table reads:

Manufactured textures - rolled, embossed, knurled, diamond-cut - have a known slope ceiling, so a fixed grazing angle clears them. The reflection law behind this table is the one traced in bright field versus dark field lighting.

For flat parts the first light to try is the LS-RNG-0256, a 96 mm OD ring light whose beam sits in the low-angle (0-15°) class, running at 24 V.

If the texture arrives defined on a drawing, note that Ra and Rz values do not convert into a slope ceiling - you need the profile itself. The surface texture specification on drawings and CAD models sets out what the symbols pin down and what they leave open. Otherwise, measure your own δmax in one evening with one light:

1. Clamp the part under the camera. Light it from one side with a single arc light at a known elevation; start at 45°.
2. Drop the elevation in 5° steps, one exposure each, gain fixed.
3. Note the highest elevation at which no glint survives. That elevation is e = 90° - (2δmax + θa).
4. Solve for the slope: glints gone below 20° elevation, with θa = 0.9°, gives δmax = (90 - 20 - 0.9) / 2 ≈ 35°.

An arc-shaped light such as the LS-RNG-0877, rated 1 W red / 1.7 W white at 260 g, is enough for the sweep. Subtract 5° of margin from your clean elevation, then buy the beam class that sits under it.

How close must a dark-field ring light sit?

Tens of millimetres, not hundreds: published set-up guidance for dark-field ring lights works around 50 mm of standoff, and the grazing geometry of a low beam explains why that number is small. A beam at 10° elevation slides sideways by the height change divided by tan(10°): lift the ring 10 mm and the lit strip walks 10 / 0.176 = 57 mm across the part, off a 40 mm field entirely. Mount the ring just above the part and let the lens look over it.

Size follows from the same geometry: the ring's outer diameter belongs between 1 and 2 times the field-of-view diagonal, so grazing rays from its rim still converge on your patch. The ring lights finder filters by size from under 50 mm up to 400 mm and beyond, and the light finder is the quickest way to shortlist by angle class.

Two more consequences of grazing geometry: the ring's inner diameter must clear your fixture and any lens that dips low, and single-ring coverage dies on curved parts, where the local surface normal swings across the field. On curved white parts, fire two or four arcs in sequence at their own elevations - the pattern in stacked ring lights switching angles - or move straight to polarization. The same low-angle reasoning behind pit detection is traced in detecting dents and pits with low-angle lighting.

What if the texture is truly random like sandpaper?

Then no fixed angle is clean: a random texture carries facets past 45°, so i > 2·45° + θa demands more than 90° of incidence, which no light can deliver, and you switch levers to crossed polarizers. Put a linear polarizer on the light and a second sheet, the analyzer, on the lens, crossed against it. The specular glint keeps its polarization state on reflection and the analyzer blocks it; the light that diffuses through the white pigmentation scatters many times, loses its state, and passes. You keep the texture signal and lose the mirror.

The price is 2 to 3 stops, a factor of 4 to 8 in light: cross polarization to reduce surface glare shows the set-up, and the cost of polarizers in f-stops measures the loss. If your unpolarized dark-field exposure is 1 ms, budget 4 to 8 ms with the sheets in place; at 20 parts per minute, one 8 ms strobe every 3 s is 0.3% duty.

The statistical alternative, when even crossed sheets cost you too much light: fire four exposures with the light rotated 90° between them and keep the darkest value at each pixel. Glints land somewhere new each time; texture does not. That is the sequential pattern in sequential lighting for multi-view inspection, and it is also the doorway to photometric stereo ring lighting when you want slope numbers, not just a clean image. The LS-SPC-0142, a Ø80 mm multi-angle ring light, steps its beam angle per exposure under controller command, which is exactly what this routine needs.

Two lights carry the random-texture workflow:

What does a complete set-up look like?

One worked example, a 40 mm by 33.6 mm patch of white textured polymer: a 2448 x 2048 monochrome camera with 3.45 µm pixels, a 35 mm lens at 200 mm working distance, f/5.6, and a 96 mm low-angle ring roughly 50 mm above the part. Optics first. The sensor is 2448 x 3.45 µm = 8.45 mm wide, so the magnification is m = 8.45 / 40 = 0.21. From 200 mm of working distance, the focal length is f = 200 x 0.21 / 1.21 = 34.7 mm: fit a 35 mm lens and the conjugate lands at WD = 35 x 1.21 / 0.21 = 201.7 mm, close enough to plan around 200 mm. One pixel then covers 3.45 / 0.21 = 16 µm on the part, and a 120 µm grit facet spans 120 / 16 = 7 pixels - enough for detection and for slope shading.

The lens focal length calculator runs the same equation on your own sensor:

Now the light. The measured facet ceiling is δmax = 35°. At m = 0.21 and f/5.6 the object-side cone is θa = arctan(0.21 / (2 x 5.6 x 1.21)) = 0.9°, so the beam must stay below e = 90 - 70 - 0.9 = 19.1° elevation - squarely the low-angle (0-15°) class. The field diagonal is √(40² + 33.6²) = 52 mm, so the ring's outer diameter belongs between 52 and 104 mm. At 96 mm OD the light sits at 1.8 times that diagonal: high in the window, inside it.

Two sanity checks close the optical budget. The Airy disc at f/5.6 and 550 nm is 2.44 x 0.55 x 5.6 = 7.5 µm on the sensor, 36 µm on the part - 2.2 pixels, so the lens resolves 120 µm facets with margin. Depth of field with a 2-pixel circle of confusion (6.9 µm) is 2 x 0.0069 x 5.6 x 1.21 / 0.21² ≈ 2.1 mm, an estimate that comfortably covers 1 mm of part height variation.

Exposure: unpolarized dark field on this white surface reaches mid-grey at a 1 ms strobe. Add the crossed polarizers and the same scene needs 4 to 8 ms. Strobe the light, gate the camera to the strobe, keep the part clamped, and motion blur never enters the image at either exposure.

When is dark field the wrong tool for a white surface?

When the surface is fully matte: pigment-loaded white with no gloss has no specular component left to steer away, so dark field only starves the sensor while a diffused on-axis ring at 90° delivers the photons. Even then, keep the light directional - a dome flattens the shading differences that carry the texture signal.

Four more cases where the standard answer misleads:

• The part is curved. The local normal swings across the field, so one elevation cannot hold at every point. Sequence arcs from several sides, or polarize.
• The field is large. Grazing coverage collapses as the field grows; past a few hundred millimetres, graduate to four arc lights or a segmented source from special lights, such as the LS-SPC-0177, a 120 x 65 mm 2.5D zone segmented light with a Ø120 mm footprint. The trade-offs are laid out in large field of view lighting.
• You need texture numbers, not an image. Slope maps and pseudo-roughness come from photometric stereo, not from any single ring.
• The glints are the signal. If the job is mapping gloss uniformity, keep the near-dark-field geometry and calibrate it. If you need print or marks on a curved shiny part rather than texture, follow the dome and polarizer routes for shiny metal pins instead.

What should I check before ordering?

Five checks: ring outer diameter between 1 and 2 times the field diagonal, lens barrel clearance inside the ring's inner diameter, polarizer film cut for both the light face and the lens, controller channels if you sequence angles, and strobe headroom for the polarizer's extra 2 to 3 stops.

• Ring size. A 52 mm diagonal field wants a 50-100 mm class ring; at 96 mm OD the ring is 1.8 times that diagonal. Too large and the grazing rays walk past the field; too small and the ring shadows its own patch.
• Mechanical clearance. The ring sits close to the part, so confirm the inner diameter passes your fixture and lens front. The LS-RNG-0730 is a 103 mm ring light with a 48 mm inner diameter and angle options from low angle (0-15°) through 50° to 85°.
• Polarizers. Sheet for the light face, mounted a few millimetres clear of the LEDs, and a rotating analyzer ring for the lens. Cross them on a glossy sample before the part arrives.
• Controller. One channel per light if you sequence angles; the 90°-rotation routine needs no hardware moves when each arc has its own channel. Strobe energy, not average current, pays the polarizer tax - the reasoning is in LED overdrive and current control.
• Prove stability. Glints that appear on every part are geometry; glints that drift between parts are ambient or thermal. Separate the two with the machine vision lighting stability test before the cell ships.

Three lights, three strategies:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

The decision, condensed:

• Manufactured texture, slope ceiling up to 30°, flat part: a low-angle dark-field ring mounted within tens of millimetres. Buy the ring first.
• Slope ceiling 30-40°: stay in the low-angle (0-15°) class and stop down to f/8 to trim the cone by a few tenths of a degree.
• Random texture like sandpaper: crossed polarizers plus whatever geometry is convenient, with 2 to 3 stops of strobe energy budgeted.
• Unknown texture: one arc light and the 5°-step sweep; measure δmax before buying the ring.

Next step: size the light in the lighting configurator, browse the light finder, or send us the part - expert support comes with every order. More lighting geometry and colour guidance sits on the lighting topic page.