When black objects look white in your images: missing IR-cut filter

A camera turns an ordinary room into a problem: the black gasket, the black felt backdrop or the black cable jacket comes out white, pastel or washed-out pale, while white objects look close to normal. Daylight frames look fine; the same scene under halogen at night collapses. That signature — lifted blacks, pastel colors, daylight apparently fine — is almost never a broken sensor and never a white balance you can tune out. It is the camera seeing near-infrared, which happens when the IR-cut filter is missing. This guide gives you the five-minute tests that confirm it, the arithmetic that explains the pastel look, the other causes ranked with a test for each, and the ordering decision for both outcomes: color inspection, where infrared must be blocked, and infrared work, where it must be let in.

Why do black objects look white in my images?

Black dye absorbs almost all visible light but passes or reflects much of the near-infrared band, so a sensor whose IR-cut filter is missing records that infrared as brightness, and under IR-rich lighting such as halogen the object renders as pale, washed-out pastel instead of black.

A silicon sensor responds from roughly 400 nm out to about 1,000-1,100 nm; your eye stops near 700 nm. The IR-cut filter — a dichroic coating on the sensor window or inside the lens — rejects that invisible band so the camera sees what you see. Remove it and the sensor gains a second channel you cannot check by eye.

"Black" is a promise about the visible band only. The dyes that make fabric read black typically reflect 2-5% of visible light but 30-60% of near-infrared, which is why infrared photography is full of black cloth that glows. The leak adds a similar pedestal to all three Bayer color channels; where the visible signal is nearly zero, the pedestal is all that remains, and the object reads bright and colorless.

The arithmetic, worked once. The ratio that decides it:

A 3,000 K halogen filament radiates roughly 10% of its watts in the 380-700 nm band and roughly 17% in the 700-1,000 nm band silicon still reads; most of the rest sits beyond 1,100 nm where silicon is blind. Take a black cloth with 4% visible reflectance and 40% near-infrared reflectance, both typical for black dyes:

• Visible return: 0.10 × 0.04 = 0.004
• Near-infrared return: 0.17 × 0.40 = 0.068
• Ratio: 0.068 / 0.004 = 17

The cloth returns about 17 times more signal through its infrared window than through its visible window. Repeat with daylight shares — roughly 45% visible, 30% in the 700-1,000 nm band — and the ratio falls to 0.30 × 0.40 / (0.45 × 0.04) = 0.12 / 0.018 ≈ 7. The lift survives daylight, just smaller and hidden beside correctly colored surroundings, which is why a daylight snapshot can pass while the same camera quietly drifts every color measurement you make. Treat these band shares and reflectances as rounded estimates: the order of magnitude is the diagnosis, not the third decimal.

Both camera kinds are sold on purpose, which is why the spectrum axis in how machine-vision cameras are classified, axis by axis matters before you buy, and why the area scan cameras finder tags deliberately infrared-sensitive models under a spectrum filter.

How do I test for a missing IR-cut filter in five minutes?

Two five-minute tests settle it: photograph the same black object in daylight and under halogen, then press a TV remote at the lens through a piece of that fabric — if daylight looks right, halogen goes pastel and the remote glows through the weave, the filter is gone.

Test one, two illuminants. Shoot the same gasket, felt backdrop or cable jacket in daylight and under a 3,000 K lamp. Blacks that lift under the lamp and hold in daylight point at infrared, because no white-balance error behaves that way: a white balance mistake shifts colors everywhere, it does not discriminate against black objects.

Test two, an infrared source. A TV remote's LED runs near 940 nm, just inside silicon's response, which reaches roughly 1,000-1,100 nm. In a dim room, an unfiltered camera shows the flashing remote as a visible glow, and the glow passes through black fabric that blocks everything your eye can see. An 850 nm LED board or indicator reads far brighter if you have one on the bench. Judge the result on the camera under test, not on a phone — if you want the phone trick for checking the emitter itself, testing board IR illuminators the phone-camera way walks through it.

Why doesn't white balance or exposure fix it?

White balance multiplies each color channel by a gain estimated from the whites in the scene, and no multiplication can subtract the infrared pedestal the black object itself is adding, so every white balance mode leaves the object pastel — cycling through all of them proves the point in ten minutes.

Each channel's output is its gain times (visible signal plus infrared leak). On white and grey objects the gains rebalance the ratios and hide much of the leak. On a black object there is no visible signal to rebalance — only the leak, and it reads as brightness. Exposure behaves the same way: shortening it scales every signal equally, so the infrared-to-visible ratio never moves. That is why the symptom survives every preset, a locked manual white balance and any exposure you choose — and why the only in-place cure is spectral: a filter that rejects the infrared before it reaches the sensor.

Lock the white balance to a fixed value first anyway; it costs nothing and rules out the second cause below. The settings that matter — exposure, gain, white balance — are the ones covered by runtime control of exposure and white balance, and none of them performs spectral subtraction.

What else turns blacks pale?

Once the filter tests come back clean, rank the remaining causes by probability: a locked or mis-guessed white balance, exposure clipping, mains flicker and, last of all, a damaged color filter array — each has a test that takes under a minute.

1. White balance locked to the wrong illuminant. Set a fixed white balance matched to the lamp type and re-shoot. If blacks return and colors correct, you are done — no hardware fault.
2. Exposure or gain clipping. Open the histogram: pixels pinned at full scale (255 in 8-bit, 4,095 in 12-bit) mean clipped highlights. Halve the exposure; if everything darkens together, the scene was simply overexposed. If blacks lift while the rest of the scene sits correctly exposed, clipping is not your cause.
3. Mains flicker. A 50 or 60 Hz lamp with a short exposure produces striped bars and frames of alternating brightness — a moving, spatial symptom rather than a uniform pale wash. The black bars in some frames: flicker vs exposure plan separates flicker from exposure in three checks.
4. Damaged color filter array or lens coating. Rare, and it stays in daylight too. Swap the lens first, then the camera body; the fault follows whichever part carries it.

Is the camera defective or the wrong variant?

Usually neither: filtered and unfiltered camera variants are both deliberate products — one built to reject near-infrared for accurate color, one built to pass it for infrared lighting — so a defective module is more often the wrong variant reaching your bench with the right-looking label.

Cameras sold for night vision or infrared illumination omit the cut filter on purpose. Cameras sold for color measurement include it. The two look identical from the outside — same housing style, same mount, same connector — and a mixed-up build or a mislabeled box delivers the wrong one. When a module sold as the filtered version behaves as the unfiltered one, treat it as a wrong-variant unit and return it; there is nothing to repair.

To tell what you actually have, read the sensor's spectral response plot in its data sheet: an IR-cut color camera's response collapses above roughly 650-700 nm, while an infrared-capable camera stays high out toward 1,000 nm. If the papers do not say, the remote test decides in the dark. If your application sits between the two — mostly color, occasional infrared — color camera near-infrared imaging: IR-cut filter options compares fixed, removable and switchable filters.

The decision: buying for color inspection, the wrong variant goes back and an IR-cut model takes its place. Buying for infrared illumination, it is the right variant — the pastel look under visible lamps is the physics doing exactly what the camera was built to do, and that job starts in the infrared cameras range.

Can an external IR-block filter save the camera?

Yes: a threaded IR-block (hot-mirror) filter in front of the lens cuts the near-infrared at its cut-on near 650-700 nm and restores normal color, at the cost of the camera's infrared sensitivity and a few percent of visible transmission — and it fits the camera you already own.

The filter is a dichroic coating on a plane-parallel disc: it passes the visible band and reflects the infrared back toward the scene. Machine-vision types thread into the rear of the lens or the camera's filter pocket, and color-accurate types cut on around 650-700 nm so the red channel passes whole. Expect exposures to lengthen by a few percent — the visible loss is that small. A filter cannot make a monochrome sensor see color; it only removes the infrared pedestal.

Which camera should I order instead?

For color inspection, order a color camera whose data sheet states IR-cut filtering — for example a 2.5 MP color model on a 1/1.7-inch sensor with global shutter, C-mount and GigE — and for deliberate infrared work, order a camera sold as NIR- or SWIR-sensitive instead of inheriting the accident.

Start in the camera finder; the system configurator's camera step walks the same questions in order. Shortlist by sensor and interface, then read each candidate's spectral response plot — that plot is where IR-cut versus infrared-sensitive is written.

A representative replacement for a general color station:

The MV-CU020-90GC is a 2.5 MP color camera on a 1/1.7-inch sensor with global shutter, C-mount and GigE — a compact, ordinary choice for bench color checks where the infrared must stay out.

When infrared is the point — 850 or 940 nm lighting, dye and moisture contrast between materials — SWIR vs NIR vs visible imaging maps the bands to the job. A representative model:

The MV-CI013-GS-NN is a 1.3 MP SWIR camera on the IMX990 sensor, 1/2-inch format, C-mount and GigE, in a fanless housing.

The decision, per case: color inspection — a filtered color camera, with a threaded IR-block filter as the backup plan; infrared work — a NIR or SWIR camera with no cut filter and infrared lighting planned in; both jobs on one station — two cameras or a removable filter slide, never one camera asked to do both.

How do I keep this from happening again?

Make spectral behavior a purchase line item instead of a bench discovery: confirm the filter type before ordering, run the halogen-and-remote test on the first camera of every new order, and treat LED lighting as a mask rather than a cure.

• At purchase: read the sensor's spectral response plot and buy the filter behavior the job needs; the spectrum filter in the camera finder separates NIR- and SWIR-capable models from color cameras.
• At purchase: add a threaded IR-block filter to the order for color stations on mixed-lit floors — the transmission penalty is a few percent.
• At incoming: one halogen desk lamp and one TV remote reproduce both five-minute tests on the bench; log the result against the camera serial before the camera reaches a station.
• In the station: white LED lighting emits little energy beyond 700 nm, so it starves the fault — but the camera stays wrong, and the next halogen lamp or the next night shift exposes it again.
• In the recipe: never use an unfiltered camera for color matching; even in daylight the infrared pedestal sits several times above the visible return, per the arithmetic above.

More camera-selection reading sits on the cameras topic page. If you are staring at a bench of pastel images right now, run the two tests first — then take the winning branch above, or send the two test images over via the contact page and the applications team will read them with you.