Cross polarizer on lens and light to kill specular glare

Specular glare off glossy plastic, glass caps, sealed packaging and oily films washes out the exact region you need to inspect. The standard cure is cross polarization: linear polarizing film over the light, a linear polarizing filter on the lens, axes set 90° apart. This guide walks through the fit-and-align procedure, the extinction test that proves the rig, the exposure arithmetic that pays the light budget back, and the red-bandpass fix for curved parts under ambient light — plus the cases where crossed polarizers are the wrong tool.

Which filter goes on the lens and which goes on the light?

Linear polarizing film goes over the light and a linear polarizer goes on the lens; machine-vision sensors have no autofocus or metering, so circular versions add nothing. The two axes sit at 90° to each other.

The film defines the polarization of the illumination; the analyzer rejects exactly that state after the part reflects it. Glare from a dielectric keeps its polarization because dielectric reflection is polarization-sensitive — completely polarized only at Brewster's angle, about 56° for glass and most plastics (arctan n), but strongly polarized at the 20-45° incidence a ring light usually produces. The geometry is at Brewster's angle.

Sheet film covers the light; a threaded filter or a film square in a rotatable holder covers the lens. Visible-band film works across roughly 400-700 nm; for near-infrared lighting at 850 or 940 nm, switch to wire-grid polarizers rated for 700-1,100 nm. The imaging physics behind the technique is covered in more depth by the guide to cross polarization for surface glare.

How much glare reduction does a crossed pair actually give?

On a flat dielectric the crossed pair leaves about 1/50th of the lamp's specular glare with standard 100:1 film, and about 1/500th with 1,000:1 film, because the leaked light is the sum of both films' crossed-axis leakage.

Ideal crossed polarizers pass nothing; real film leaks a little light through the crossed axis, described by the extinction ratio. Polarizer selection guides tabulate sheet-film extinction from below 100:1 for economical film to 10⁶:1 and above for high-purity optics. What reaches the sensor as residual glare is the sum of the two leaks:

With 100:1 film on both sides, ε = 0.01 each, so the reduction is 1 / 0.02 = 50:1. With 1,000:1-class film it is 1 / 0.002 = 500:1. Rotation δ away from extinction adds a leak of its own through sin²δ, and curved or internally scattering parts depolarize the glare, so the scene-level number always lands below the film pair's ceiling:

Treat 50:1 with budget film, a few hundred with good film, as the flat-dielectric ceiling — then measure what your part actually gives.

How do you fit the film and align the pair?

Tape the film taut over the diffuser without wrinkles, rotate the lens polarizer from brightest to black, then trim to within 2° of extinction — at 5° off, misalignment alone caps the reduction near 130:1.

1. Cut the film at least 20 mm larger than the light's illuminated face on every side and tape it taut across the diffuser or cover glass. A wrinkle rotates the polarization locally and leaks glare as a bright streak.
2. Mount the analyzer: a threaded filter screws into the lens front thread; a film square goes into a rotatable holder when the lens has no thread, as with most board lenses.
3. Lock exposure and gain first so nothing on the camera compensates while you rotate.
4. Rotate the analyzer through a full turn and watch the lamp's reflection in the part. Brightness peaks twice, 180° apart, and extinction sits exactly halfway between the two bright positions — that centers you within about a degree before any fine trim.
5. Trim to the darkest reflection and mark the holder against a fixed reference so the angle survives cleaning and bumps.

How do you measure extinction and prove the rig works?

Shoot the part with the analyzer parallel and then crossed at locked exposure: a peak ratio of 20:1 or better means the rig is right, and the crossed specular peak should sit within about 10 DN of the diffuse background.

Auto exposure and auto gain hide exactly what you are measuring, so lock both; the trade-offs are in the guide to automatic iris, exposure and gain control. Then image the shiny dielectric you actually inspect — the part itself, or a glass slide as a stand-in — and read the peak pixel value in the glare region at both analyzer positions. Illustrative numbers from a cap cell: parallel peak 240 DN, crossed peak 9 DN, so the scene reduction is 240 / 9 = 27:1.

Above 20:1, the rig is right: film flat, edges masked, alignment within a couple of degrees. Below 10:1, work down the list: wrinkled film, lamp light bypassing the film edges, a metal region inside the field, or film whose extinction is specified below 100:1. On a good rig the crossed specular peak also sits within about 10 DN of the diffuse background — the highlight gone, not merely dimmed, which is what the inspection algorithm needs. Re-run the peak test at each shift start; it takes a minute and catches a knocked holder before it costs a batch.

How much light do the two polarizers cost, and what exposure buys it back?

Two film layers pass roughly 40% each, so the sensor sees 0.42 × 0.42 ≈ 18% of the light — a 5.7× loss, about 2.5 stops — which you repay with two stops of aperture and 1.5× more exposure, or with a harder strobe pulse.

Worked example, geometry and exposure in one pass. The part is a 40 mm sealed cap on a line moving at 20 mm/s. The camera is a 2/3" sensor, 8.45 mm wide, 2448 pixels of 3.45 µm (illustrative), and the target field is 52 mm at a 250 mm working distance:

• Magnification: m = 8.45 mm / 52 mm = 0.163
• Focal length: f = WD × m / (1 + m) = 250 × 0.163 / 1.163 = 35.0 mm
• Check with a 35 mm lens: m = 35 / (250 − 35) = 0.163, field = 8.45 / 0.163 = 51.9 mm
• Pixel scale at the part: 3.45 µm / 0.163 = 21.2 µm per pixel

Without polarizers, 2 ms at f/4 puts the cap's white face near 200 DN. The film pair multiplies the light by 0.42 × 0.42 = 0.176, so holding 200 DN takes 2 ms × 5.7 = 11.4 ms at f/4. At 20 mm/s the part smears 20 mm/s × 11.4 ms = 0.23 mm, about 11 pixels — unusable. Open two stops instead, f/4 to f/2, and the exposure lands at 2 ms × 5.7 / 4 = 2.9 ms, a smear of 0.058 mm or about 3 pixels — fine for presence checks, marginal for edge measurement. On a moving line the clean fix is a strobe: a 300 µs pulse freezes motion to 20 mm/s × 0.3 ms = 6 µm, under a third of a pixel, with the controller making up the 5.7× inside the pulse.

So: strobe on moving parts and keep the lens near f/4; on a static station, f/2 with a 2.9 ms exposure is simpler, at the cost of roughly half the depth of field.

The geometry above lands on a 35 mm C-mount lens in the 6 MP class, and the LS-FFL-0679 is that lens.

The polarizer light-loss guide runs the f-stop accounting in general, and the LED strobe duty cycle guide shows how to size pulse and current together. If the cell fixes the working distance instead of the field, pick the focal length to match in the machine vision lenses category.

Why does glare survive on curved parts under ambient light?

Ambient light is unpolarized, so the analyzer passes about 40% of it whatever the alignment; the fix is a narrowband red lamp near 620 nm with a matched bandpass on the lens, which cuts out-of-band ambient by three or more orders of magnitude.

The analyzer sorts by polarization, not by origin. Room and sunlight are unpolarized before and after reflection, so ambient glare loses only the filter transmission and re-adds brightness across the whole curved surface. Curvature also rotates the lamp's own polarization: where the local reflection turns the polarization 45° from the analyzer, Malus's law leaves cos²45° = 0.5, so half the lamp glare passes — the bright bands near the silhouette of a glossy cylinder are geometry, not bad alignment.

To make polarization sorting apply to everything in the image, make everything lamp light. Drive a narrowband red lamp near 620 nm and fit a matched red bandpass, of order 10 nm FWHM, on the lens: out-of-band ambient falls by three or more orders of magnitude (typical for narrow bandpass filters), and the small in-band remainder is dominated by lamp glare, which the crossed pair then cuts by its 50:1. The reasoning behind single-wavelength illumination is in the monochromatic lighting guide, and acceptance-test numbers are in the ambient light rejection guide.

A 120 mm ring with a 60° emission angle at 620 nm on 24 V is exactly this kind of source:

Film over its diffuser completes the polarized, narrowband lamp; center the bandpass on the lamp wavelength and re-run the peak test, because film transmission varies with wavelength.

When does cross polarization fail or make things worse?

Cross polarization works on dielectrics — plastic, glass, paint, oil, flux — and barely at all on bare metal, which mixes polarization states on reflection; stressed plastic leaks bright patches, and a defect that is itself specular gets harder to see.

• Bare metal is where the pair fails: go diffuse instead — the dome, diffuse and polarizer guide for shiny metal and the dome lights guide compare the options, and the dome lights category is where that hardware lives.
• Stressed plastic depolarizes by birefringence and shows bright patches that no alignment removes. If stress is the inspection target, that leak is the measurement; if not, budget for the patches.
• Matte, foamed or paper-like surfaces have no specular component to kill, so the pair only costs you the 5.7× light.
• A scratch in glass or contamination on shiny film is itself specular: killing glare hides the defect. Use dark-field geometry so the defect scatters light toward the lens while the flat surface reflects it away.

What do you order and check before building the rig?

Order a ring light about twice the field of view in outer diameter, film specified above 1,000:1 extinction, an analyzer with clear aperture over the sensor's image circle and, under ambient light, the narrowband lamp-plus-bandpass pair.

• Ring light with outer diameter about twice the field of view, so the film-covered face illuminates from outside the image cone: a 100 mm ring serves the 52 mm example field.
• Linear polarizing film specified at 1,000:1 or better; economical sheet film below 100:1 caps the whole rig near 50:1 before the part even enters. Buy two sheets — film scratches and creases are not repairable.
• Analyzer clear aperture over the image circle, 11 mm on the 2/3" example sensor; any 25 mm threaded filter covers it. Confirm your lens front-thread diameter before ordering.
• A rotation holder that holds its angle under vibration, or index marks on the barrel.
• Under ambient light: narrowband lamp and bandpass ordered together so the center wavelengths match.
• A strobe-capable 24 V light controller, and re-run the lighting stability test whenever you change lamp, film or exposure.

A 100 mm diffused ring at 30° on 24 V is the example cell's light: a soft field the film covers flatly.

Browse the ring lights category or the ring light finder to filter by size, angle and wavelength, with cables and mounting parts on the same shelf. The lighting configurator steps through lamp-controller pairing, more lighting guides sit in the lighting topic hub, and you can reach us through the contact page to size the pair for your part.